Implement AVX-512 SIMD Optimizations

[kevinamccorvon]

AVX2 is working great (42% speedup on ML-DSA), but newer CPUs have AVX-512 which should give more boost.

Targets for AVX-512:

• ML-DSA (all variants)
• Hash functions: SHA-2, SHA-3, BLAKE2/3
• Elliptic curves: field arithmetic, point multiplication
• AES (if there's gains to be had)

Requirements:

• Runtime CPU detection (can't assume everyone has AVX-512)
• Benchmark against AVX2 baseline
• Ensure graceful fallback to AVX2/scalar
• Document which CPUs benefit and by how much