Add reproducible x86 assembly checks

[falseywinchnet]

Motivation

• Provide a reproducible, documented way to emit annotated x86 assembly for both AVX2/FMA and SSE2/no-AVX builds so the SSE/AVX pathways can be inspected and audited automatically.
• Record findings from the assembly audit and related micro-benchmark spot checks to guide future SIMD/locality work and to avoid regressions.

Description

• Add a separate make asm-check target that probes the compiler for AVX2/FMA and SSE2/no-AVX support and emits annotated assembly into build/src/bfft_avx2.s and build/src/bfft_sse2.s when available by using $(AVX2_FLAGS) and $(SSE2_FLAGS) in the Makefile.
• Wire assembly outputs into the build tree without changing the default make/make test flow so normal builds continue to use the automatic SIMD policy.
• Update src/ASSEMBLY_NOTES.txt with audit findings and benchmark spot checks describing AVX2/SSE2 shapes, a rejected SSE fusion experiment (high register pressure), and example benchmark numbers observed during inspection.

Testing

• Ran make clean, make, and make test with the automatic SIMD selection and all library correctness tests passed.
• Ran make asm-check which produced build/src/bfft_avx2.s and build/src/bfft_sse2.s on a host compiler that accepted those switches.
• Ran the benchmark and probe programs used for the audit: ./build/examples/benchmark 4096 1000, ./build/examples/benchmark 65536 200, and ./build/examples/locality_probe 65536 200 64; these completed and produced the recorded spot-check timings (FFTW columns were unavailable because FFTW/FFTWf were not found by dlopen).
• Built and ran an SSE2/no-AVX correctness build and its benchmark to validate the SSE pathway (g++ -msse2 -mno-avx compile); correctness tests passed for the SSE2 build.

---

Codex Task