fix(ir): integral-float counts + range-checked value-param binds (0083)

Item 2 (Agra ruling): a compile-time INTEGRAL float (`4.0`, `N : f64 :
4.0`, `N :: 4.0`) used as an array dimension / Vector lane / generic
value-param count / `inline for` bound now folds to its integer at the
shared leaf — `program_index.floatToIntExact`, used by both the
`.float_literal` arm of `evalConstIntExpr` and `moduleConstInt`. All four
consumers route through the one evaluator, so `[4.0]s64` lays out the same
`[4]s64` uniformly; a non-integral (`4.5`) or negative value stays
rejected by the downstream `foldDimU32` gate. Pass-0 now pre-registers
float-valued module consts for forward-alias parity with int consts.

Item 1: a generic value-param bind (`Box($K: u32)`) never range-checked
the folded arg, so `Box(5_000_000_000)` compiled and ran. The bind now
range-checks against the param's declared type — a `u32` count through the
shared `foldDimU32` gate (making program_index's "single u32 gate for
value-param counts" doc true), any other integer type through the new
`program_index.intTypeRange` — and emits a clean "value N does not fit in
u32 parameter K" otherwise. The declared type is threaded via a new
`TemplateParam.value_type`.

Regressions: examples 0145 (integral-float array dim), 1504 (Vector lane),
0611 (inline-for bound), 0209 (value-param integral-float), 1132
(non-integral float dim rejected), 1133 (negative float dim rejected),
1134 (oversized u32 value-param rejected) + program_index float-fold unit
tests. Gate: zig build, zig build test, 406/0 run_examples.

examples/0145-types-integral-float-array-dim.sx

@@ -0,0 +1,29 @@
+// An array dimension accepts any compile-time numeric constant whose value is a
+// positive INTEGRAL number — an integral float (`4.0`) folds to its integer just
+// like `4`. A float-typed const (`N : f64 : 4.0`), an untyped-float const
+// (`M :: 4.0`), and a direct float literal (`[4.0]s64`) all lay out the same
+// `[4]s64` as the integer spelling, so element store/read is in bounds.
+//
+// Regression (issue 0083 / F0.4 attempt 8, Agra ruling): an integral float used
+// as a dimension was wrongly rejected "must be a compile-time integer constant".
+// The shared const-int evaluator now folds an integral float literal (and a
+// float-typed module const) via `program_index.floatToIntExact`; a non-integral
+// float (`4.5`) is still rejected (see 1132).
+#import "modules/std.sx";
+
+N : f64 : 4.0;   // float-typed const
+M :: 4.0;        // untyped float const
+
+main :: () {
+    a : [N]s64 = ---;          // dim from a float-typed const
+    a[0] = 10; a[3] = 40;
+    print("a len={} a0={} a3={}\n", a.len, a[0], a[3]);
+
+    b : [M]s64 = ---;          // dim from an untyped float const
+    b[1] = 21;
+    print("b len={} b1={}\n", b.len, b[1]);
+
+    c : [4.0]s64 = ---;        // direct integral-float-literal dim
+    c[2] = 32;
+    print("c len={} c2={}\n", c.len, c[2]);
+}

examples/0209-generics-value-param-integral-float.sx

@@ -0,0 +1,19 @@
+// A generic value parameter (`$K: u32`) binds a literal (`Vec(3, s64)`) and an
+// integral-float named const (`Vec(L, s64)` with `L : f64 : 4.0`) to the same
+// integer a plain `4` would — the value-param arg folds through the shared
+// const-int evaluator, so the integral-float rule (F0.4 attempt 8, Agra ruling)
+// reaches value params too. The folded value is the array length `[K]s64`.
+//
+// The bind is range-checked against the declared `u32` (an out-of-range arg is a
+// clean compile error — see 1134); a valid in-range value binds normally.
+#import "modules/std.sx";
+
+Vec :: struct ($K: u32, $T: Type) { data: [K]T; }
+
+L : f64 : 4.0;
+
+main :: () {
+    a : Vec(3, s64) = ---;       // literal value param
+    b : Vec(L, s64) = ---;       // integral-float named-const value param → 4
+    print("a.len={} b.len={}\n", a.data.len, b.data.len);   // 3 and 4
+}

examples/0611-comptime-integral-float-inline-for.sx

@@ -0,0 +1,14 @@
+// An `inline for 0..M` bound accepts an integral float constant — `M :: 3.0`
+// unrolls the same three iterations as `M :: 3`. The inline-for bound folder
+// (`evalComptimeInt`) delegates to the shared const-int evaluator, so the
+// integral-float rule (issue 0083 / F0.4 attempt 8, Agra ruling) applies here
+// too.
+#import "modules/std.sx";
+
+M :: 3.0;
+
+main :: () {
+    s := 0;
+    inline for 0..M: (i) { s += i; }
+    print("sum 0..M = {}\n", s);          // 0 + 1 + 2 = 3
+}

examples/1132-diagnostics-array-dim-non-integral-float.sx

@@ -0,0 +1,14 @@
+// A NON-integral float constant (`4.5`) used as an array dimension is a hard
+// error — only an integral float (`4.0`) folds to a count. Clean diagnostic +
+// non-zero exit, NOT a fabricated length.
+//
+// Regression (F0.4 attempt 8, Agra ruling): the integral-float rule accepts
+// `4.0` as a dimension but must keep rejecting `4.5` (it is not an integer).
+#import "modules/std.sx";
+
+N : f64 : 4.5;
+
+main :: () {
+    a : [N]s64 = ---;
+    print("unreachable: {}\n", a.len);
+}

examples/1133-diagnostics-array-dim-negative-float.sx

@@ -0,0 +1,12 @@
+// A NEGATIVE integral float (`-2.0`) used as an array dimension is a hard error.
+// The integral-float rule folds and negates it to `-2`, then the shared u32 dim
+// gate rejects a below-minimum dimension — a clean diagnostic + non-zero exit.
+//
+// Regression (F0.4 attempt 8, Agra ruling): integral floats fold, but a negative
+// result is still rejected (a dimension must be non-negative).
+#import "modules/std.sx";
+
+main :: () {
+    a : [-2.0]s64 = ---;
+    print("unreachable: {}\n", a.len);
+}

examples/1134-diagnostics-value-param-u32-overflow.sx

@@ -0,0 +1,17 @@
+// A generic value-param arg that does not fit the param's declared integer type
+// (`Box(5_000_000_000)` for `$K: u32`) is a hard error — a clean diagnostic +
+// non-zero exit, NOT a silent truncating bind.
+//
+// Regression (F0.4 attempt 8, item 1): `resolveValueParamArg` bound the folded
+// i64 without range-checking the declared type, so an out-of-u32 arg compiled
+// and ran. The bind now routes a `u32` count through the shared
+// `program_index.foldDimU32` gate (the same one array dims / Vector lanes use),
+// so an oversized value is rejected before instantiation.
+#import "modules/std.sx";
+
+Box :: struct ($K: u32) { value: s64; }
+
+main :: () {
+    b : Box(5000000000) = ---;
+    print("unreachable\n");
+}

examples/1504-vectors-integral-float-lane.sx

@@ -0,0 +1,12 @@
+// A Vector lane count accepts an integral float constant — `L : f64 : 4.0` lays
+// out the same `Vector(4, f32)` as the literal `4`. The lane resolver shares the
+// const-int evaluator with the array-dim path, so the integral-float rule
+// (issue 0083 / F0.4 attempt 8, Agra ruling) applies uniformly.
+#import "modules/std.sx";
+
+L : f64 : 4.0;
+
+main :: () {
+    v : Vector(L, f32) = .[1.0, 2.0, 3.0, 4.0];
+    print("v0={} v2={} v3={}\n", v[0], v[2], v[3]);
+}

examples/expected/0145-types-integral-float-array-dim.exit

@@ -0,0 +1 @@
+0

examples/expected/0145-types-integral-float-array-dim.stderr

@@ -0,0 +1 @@
+

examples/expected/0145-types-integral-float-array-dim.stdout

@@ -0,0 +1,3 @@
+a len=4 a0=10 a3=40
+b len=4 b1=21
+c len=4 c2=32

examples/expected/0209-generics-value-param-integral-float.exit

@@ -0,0 +1 @@
+0

examples/expected/0209-generics-value-param-integral-float.stderr

@@ -0,0 +1 @@
+

examples/expected/0209-generics-value-param-integral-float.stdout

@@ -0,0 +1 @@
+a.len=3 b.len=4

examples/expected/0611-comptime-integral-float-inline-for.exit

@@ -0,0 +1 @@
+0

examples/expected/0611-comptime-integral-float-inline-for.stderr

@@ -0,0 +1 @@
+

examples/expected/0611-comptime-integral-float-inline-for.stdout

@@ -0,0 +1 @@
+sum 0..M = 3

examples/expected/1132-diagnostics-array-dim-non-integral-float.exit

@@ -0,0 +1 @@
+1

examples/expected/1132-diagnostics-array-dim-non-integral-float.stderr

@@ -0,0 +1,5 @@
+error: array dimension must be a compile-time integer constant
+  --> examples/1132-diagnostics-array-dim-non-integral-float.sx:12:10
+   |
+12 |     a : [N]s64 = ---;
+   |          ^

examples/expected/1132-diagnostics-array-dim-non-integral-float.stdout

@@ -0,0 +1 @@
+

examples/expected/1133-diagnostics-array-dim-negative-float.exit

@@ -0,0 +1 @@
+1

examples/expected/1133-diagnostics-array-dim-negative-float.stderr

@@ -0,0 +1,5 @@
+error: array dimension must be non-negative, got -2
+  --> examples/1133-diagnostics-array-dim-negative-float.sx:10:10
+   |
+10 |     a : [-2.0]s64 = ---;
+   |          ^^^^

examples/expected/1133-diagnostics-array-dim-negative-float.stdout

@@ -0,0 +1 @@
+

examples/expected/1134-diagnostics-value-param-u32-overflow.exit

@@ -0,0 +1 @@
+1

examples/expected/1134-diagnostics-value-param-u32-overflow.stderr

@@ -0,0 +1,5 @@
+error: value 5000000000 does not fit in u32 parameter K
+  --> examples/1134-diagnostics-value-param-u32-overflow.sx:15:13
+   |
+15 |     b : Box(5000000000) = ---;
+   |             ^^^^^^^^^^

examples/expected/1134-diagnostics-value-param-u32-overflow.stdout

@@ -0,0 +1 @@
+

examples/expected/1504-vectors-integral-float-lane.exit

@@ -0,0 +1 @@
+0

examples/expected/1504-vectors-integral-float-lane.stderr

@@ -0,0 +1 @@
+

examples/expected/1504-vectors-integral-float-lane.stdout

@@ -0,0 +1 @@
+v0=1.000000 v2=3.000000 v3=4.000000