fix(fuzz): carve out return-value-placement dead stores — closes #112

[avrabe]

Summary

Closes #112 — the exploration fuzz harness `i64_lowering_doesnt_clobber_params` was false-positiving on the `Movw R0, 0; Mov R0, R8` return-value-placement pair. The Movw zeros R0 and is then immediately overwritten by the Mov — a redundant dead store, not a real AAPCS clobber. The param IS already preserved at that point in the function.

This is the option (b) harness-side fix per the issue's investigation comment, deliberately cheaper than the underlying codegen peephole (option a, deferred to a future codegen-quality PR).

Carve-out

For each ARM instruction in the param-protection window, check the immediately-following instruction: if it also writes the same param reg, the current write is provably dead and gets skipped. Indexed iteration via `arm_instrs.get(instr_idx + 1)` keeps the lookup O(1) and side-steps the prior `std::ptr::eq` hack-shape.

Soundness

The carve-out is local — it only suppresses writes whose result is provably dead at the next instruction. A real mid-computation clobber would either:

• have a non-overwriting next-op → still flagged.
• be part of a chain where the final write overwrites R{p} → still flagged on the final write (its next op is typically Pop, which doesn't write the param reg).

So the carve-out does not hide real AAPCS clobbers; it only suppresses the duplicate report on the first write of a write-then-overwrite pair.

What this enables

Once green for a couple of fuzz-smoke cycles, this harness can be promoted from `gating: false` to `gating: true` in `.github/workflows/fuzz-smoke.yml` — tracked as a follow-up.

Test plan

• CI green
• Fuzz-smoke `i64_lowering_doesnt_clobber_params` no longer reports on the `Movw R0, 0; Mov R0, _` pattern
• No regression in the LocalSet/LocalTee carve-out path (existing logic preserved verbatim)

Refs: #112 (will be auto-closed by this PR's merge).

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!