# 0178 — protocol impl method with a mismatched return/param TYPE silently miscompiles

## Symptom

An `impl P for T` whose method has the right NAME but a mismatched return type or
parameter type is accepted (it satisfies the issue-0176 conformance gate, which
is name-based), and dispatch through the erased protocol silently produces the
WRONG result (exit 0). No diagnostic. (Arity mismatch and `#builtin`-body
mismatch fail loudly — exit 1 — and are not this bug; the TYPE-mismatch cases are
silent.)

## Reproduction

```sx
#import "modules/std.sx";
P :: protocol { val :: (self: *Self) -> i64; }
T :: struct { n: i64 = 7; }
impl P for T { val :: (self: *T) -> bool { return true; } }  // return type bool ≠ i64
main :: () {
  t := T.{ n = 7 };
  p : P = t;
  print("{}\n", p.val());   // prints "1" (the bool), silently wrong — no diagnostic
}
```

A parameter-type mismatch (`x: bool` where the protocol declares `x: i64`)
similarly dispatches silently wrong.

## Investigation prompt

The issue-0176 conformance gate (`firstUnimplementedMethod` in
`src/ir/lower/protocol.zig`) checks method PRESENCE (and rejects `type_params >
0`), but does NOT check that the impl method's SIGNATURE (parameter types,
arity, return type) matches the protocol method's declared signature. A
mismatched-type impl builds a thunk that calls the impl with the wrong ABI,
silently miscompiling. Add signature validation when registering / gating an
impl method against its protocol method: compare the impl method's params
(after the erased `self`) and return type against the protocol declaration, and
emit a located diagnostic on mismatch (arity, param type, or return type). The
protocol method declaration is in `protocol_decl_map`; the impl FnDecl is in
`fn_ast_map`. Decide whether this lives in the conformance gate or in
`ProtocolResolver.registerImplBlock` (`src/ir/protocols.zig`). Follow the
no-silent-fallback rule. Verify: the repro is now a clean diagnostic (exit 1);
a correctly-typed impl still works; add an `examples/diagnostics/11xx-...`
negative regression. (Found during adversarial review of issue 0176.)