ir: fix tuple literal element widths (construction was garbage)

A tuple_init's element values must match its field types exactly — LLVM
`insertvalue` does no implicit conversion. An inferred `pair := (40, 2)`
lowered its elements under the enclosing fn's `target_type` (e.g. main's
s32 return), producing i32 values, while the field types were inferred
independently as s64. The {i64,i64} aggregate was filled with i32
constants, so reading any element back returned garbage (40 + 2^32) and
tuple equality was always false.

lowerTupleLiteral now lowers each element under its resolved field type
(the contextual target tuple's fields when present, else per-element
inference) and coerces to it, so value width always matches field width.
Assignment to a tuple-typed field/element now also propagates the target
tuple type. Adds examples/190-tuple-values.sx as a regression test and
examples/probes/tuple-baseline.sx as the Step 0.4 audit artifact.

examples/probes/tuple-baseline.sx

@@ -0,0 +1,62 @@
+// Feature 1 / Step 0.4 — tuple baseline probe (gate for Decision 2:
+// "tuples are first-class for pack storage").
+//
+// This file is the audit artifact: it exercises every tuple operation the
+// canonical `Combined`/`map` body relies on, EXCEPT the two pack/tuple
+// projection+spread sugars that are themselves Feature 1 work (documented
+// as gaps at the bottom). Everything here RUNS today.
+//
+// Run:  ./zig-out/bin/sx run examples/probes/tuple-baseline.sx
+
+#import "modules/std.sx";
+
+Listenable :: struct { value: s64; }      // stand-in element struct
+Combined   :: struct { sources: (s32, s32); }  // tuple-typed field (Decision 2)
+
+swap :: (a: s64, b: s64) -> (s64, s64) { (b, a); }
+fst  :: (t: (s64, s64)) -> s64 { t.0; }
+
+main :: () -> s32 {
+    // ── Block A — primitives (WORKS) ───────────────────────────────
+    pair := (40, 2);                       // inferred positional
+    print("A.idx     {} {}\n", pair.0, pair.1);
+    named := (x: 10, y: 20);               // named + numeric access
+    print("A.named   {} {} {}\n", named.x, named.0, named.1);
+    one := (42,);                          // 1-tuple
+    print("A.one     {}\n", one.0);
+    a : s64 = pair.0;                      // element into typed local
+    print("A.local   {}\n", a);
+
+    // ── Block B — storage in a struct field (WORKS; core of Decision 2)
+    c : Combined = ---;
+    c.sources = (7, 9);                    // assign tuple value to field
+    print("B.field   {} {}\n", c.sources.0, c.sources.1);
+
+    // ── Block C — return / pass / operators (WORKS) ────────────────
+    s := swap(1, 2);
+    print("C.ret     {} {}\n", s.0, s.1);
+    print("C.pass    {}\n", fst((11, 22)));
+    print("C.eq      {}\n", (1, 2) == (1, 2));
+    cc := (1, 2) + (3, 4);
+    print("C.concat  {} {}\n", cc.0, cc.3);
+    print("C.mem     {}\n", 3 in (1, 2, 3));
+    0;
+}
+
+// ── GAPS (Feature 1 work — intentionally NOT exercised above) ──────
+//
+//   G1. Tuple field projection across elements:
+//         t := (Listenable.{value=1}, Listenable.{value=2});
+//         v := t.value;        // expected: (1, 2)   — Decision 3 "tuple.field"
+//       Today: `error: field 'value' not found on type 'tuple'`.
+//       Needed by canonical `self.sources.value`.
+//
+//   G2. Tuple spread into call args:
+//         p := (10, 20);
+//         add(..p);            // expected: add(10, 20) — Decision 3 "..tuple"
+//       Today: lowers to one `undef` arg → LLVM arity verification failure.
+//       Needed by canonical `mapper(..sources.value)` and `(..sources)`.
+//
+// Both are already scheduled: parsing in Phase 1.2 (PackExpansion node covers
+// `(..pack)` / `..pack.field`), sema in Phase 2.3 ("tuple-spread parallels").
+// No separate Feature 1.5 needed — see Step 0.4 triage in CHECKPOINT-LANG.md.