[Perf] Implement x86 peephole and strength-reduction optimizations

[Simonsbs]

Goal

Reduce backend overhead and unlock runtime gains where GCC currently matches/exceeds us.

Scope

• Implement x86 peephole pass for common patterns (redundant moves, zeroing idioms, simple fusion).
• Add strength-reduction patterns for multiply/shift/add forms where valid.
• Add before/after asm snapshots for benchmark kernels.

Acceptance

• At least 3 measurable codegen wins on benchmark kernels.
• No correctness regressions in make test and perf gates.
• Runtime geometric mean improves by >=5% on expanded kernel suite.