Migrate x86 SSE2 SIMD to ARM64 NEON

[Copilot]

All compute-hot paths used x86-only SSE2 intrinsics (<immintrin.h>), making the codebase non-portable to ARM64.

Changes

• All 5 compute files (matrix_operations, hash_operations, string_search, memory_operations, polynomial_eval): wrapped existing SSE2 paths in #ifdef __x86_64__, added #elif defined(__aarch64__) NEON paths, and scalar fallbacks
• <immintrin.h> → <arm_neon.h> under aarch64 guard
• main.cpp: added __aarch64__ NEON detection branch
• Docker: ubuntu:22.04 already multi-arch; updated comment only

Bug fixes (discovered during migration)

• hash_operations.cpp: _mm_extract_epi16 was called with a runtime-variable lane index — illegal on NEON (lane indices must be compile-time constants). Replaced with vst1q_u8 into a local byte array.
• string_search.cpp: no direct NEON equivalent for _mm_movemask_epi8; added neon_movemask_epi8() helper.

Example pattern applied across all files

// Before
#include <immintrin.h>
__m128d a = _mm_loadu_pd(ptr);
__m128d b = _mm_mul_pd(a, scale);

// After
#ifdef __x86_64__
  #include <immintrin.h>
  __m128d a = _mm_loadu_pd(ptr); __m128d b = _mm_mul_pd(a, scale);
#elif defined(__aarch64__)
  #include <arm_neon.h>
  float64x2_t a = vld1q_f64(ptr); float64x2_t b = vmulq_f64(a, scale);
#else
  // scalar fallback
#endif