feat: turboquant encoding for vectors #7167

Performance Gate Passed

✅ 1106 untouched benchmarks
🆕 14 new benchmarks
⏩ 1522 skipped benchmarks¹

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
🆕	Simulation	`turboquant_compress_dim128_2bit`	N/A	15.2 ms	N/A
🆕	Simulation	`turboquant_compress_dim128_4bit`	N/A	16.1 ms	N/A
🆕	Simulation	`turboquant_decompress_dim768_2bit`	N/A	104.6 ms	N/A
🆕	Simulation	`turboquant_decompress_dim128_4bit`	N/A	10.3 ms	N/A
🆕	Simulation	`turboquant_compress_dim768_2bit`	N/A	143 ms	N/A
🆕	Simulation	`turboquant_decompress_dim128_2bit`	N/A	10.3 ms	N/A
🆕	Simulation	`turboquant_decompress_dim1536_2bit`	N/A	226.5 ms	N/A
🆕	Simulation	`turboquant_compress_dim1024_2bit`	N/A	144.4 ms	N/A
🆕	Simulation	`turboquant_decompress_dim1536_4bit`	N/A	226.8 ms	N/A
🆕	Simulation	`turboquant_decompress_dim1024_4bit`	N/A	105.2 ms	N/A
🆕	Simulation	`turboquant_compress_dim1536_4bit`	N/A	319.8 ms	N/A
🆕	Simulation	`turboquant_compress_dim1536_2bit`	N/A	307.1 ms	N/A
🆕	Simulation	`turboquant_compress_dim1024_4bit`	N/A	151.3 ms	N/A
🆕	Simulation	`turboquant_decompress_dim1024_2bit`	N/A	105.1 ms	N/A

_{Comparing claude/admiring-lichterman (de598ad) with develop (fa6931a)}

1522 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. ↩