encodings/fsst: add OptimizedDecompressor for throughput

[joseph-isaacs]

Summary

• Adds OptimizedDecompressor in encodings/fsst/src/decompressor.rs, replacing the default fsst-rs decompressor in the canonicalization path
• Processes compressed codes in 32-code blocks (4×8-byte chunks) using SWAR escape detection — checks all 8 bytes of a u64 for 0xFF in a single branchless operation
• Pre-converts the symbol table to u64 at construction time (~2.3KB total, fits in L1 cache), eliminating Symbol::to_u64() per lookup
• Re-enters the 32-code fast path after each escape rather than permanently dropping to scalar processing
• Runtime dispatch to a BMI1/BMI2/POPCNT-optimized codepath on x86-64

Test plan

• Existing unit tests in decompressor.rs cover: basic decompression, escape codes, empty input, matches baseline on random strings, matches baseline on full-byte-range inputs with many escapes, large corpus correctness
• cargo test -p vortex-fsst
• cargo bench -p vortex-fsst --bench fsst_compress -- decompress_fsst to observe throughput improvement

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will improve performance by 19.89%

⚡ 9 improved benchmarks
✅ 1154 untouched benchmarks
⏩ 1457 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	take_map[(0.1, 0.5)]	1,154.5 µs	990.4 µs	+16.57%
⚡	Simulation	take_map[(0.1, 1.0)]	2 ms	1.7 ms	+19.08%
⚡	Simulation	patched_take_10k_contiguous_not_patches	258.4 µs	228.1 µs	+13.29%
⚡	Simulation	patched_take_10k_first_chunk_only	302 µs	271.8 µs	+11.14%
⚡	Simulation	patched_take_10k_dispersed	316 µs	285.8 µs	+10.58%
⚡	Simulation	patched_take_10k_random	270.3 µs	240 µs	+12.64%
⚡	Simulation	patched_take_10k_contiguous_patches	258.1 µs	227.7 µs	+13.32%
⚡	Simulation	take_10k_first_chunk_only	270.6 µs	225.7 µs	+19.89%
⚡	Simulation	take_10k_dispersed	284.4 µs	239.5 µs	+18.76%

---

_{Comparing ji/fsst-optimized-decompressor (c1f032a) with develop (12f63a4)}

Footnotes

1. 1457 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[robert3005]

if this ends up being a perf win we should migrate this to fsst-rs imho