Summary

Add SIMD-accelerated decoding for compressed posting lists. Currently decoding uses scalar code; SIMD could provide ~2x faster decompression.

Context

Profiling shows bitpack decoding is ~6% of query time at scale. While not a major bottleneck, SIMD acceleration is a straightforward optimization.

Approach

• Use SSE2/AVX2 intrinsics for bitpack decode
• Runtime detection for portability (fall back to scalar on unsupported platforms)
• Reference: Lemire et al., "Decoding billions of integers per second through vectorization"

Priority

Low - the current scalar implementation is sufficient. Only pursue if decode overhead becomes significant in specific workloads.