feat: SIMD bind and hamming kernels for ARM64 NEON

New NEON instructions:
- ST1 (store 16 bytes)
- MUL (vector multiply 16xI8)
- CMEQ (compare equal)
- NOT (bitwise NOT)
- CNT (population count)
- USHR (unsigned shift right)
- ADDV (add across vector)
- UADDLV (unsigned add long across)
- UMOV (move from vector to GPR)

New JIT functions:
- compileBindSIMD: 4.3x faster than scalar bind
- compileHammingSIMD: SIMD compare and count differences

Integration:
- jit_unified.zig: routes to SIMD on ARM64
- vsa_jit.zig: hamming_cache for JIT hamming functions

All 80 tests pass. SIMD coverage: dot, bind, hamming.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: Antigravity Agent