Optimize ZSTD compression level 6 parameters

[BiplabRaut]

• Reduce minMatch from 3 to 2 and searchLog from 4 to 2 for default parameters
• Reduce searchLog from 4 to 3 for 16KB dictionary parameters
• Update corresponding test expectations for determinism tests

These parameter adjustments improve compression efficiency for level 6 while maintaining the lazy search strategy.

Performance Results

Test Setup: AMD Genoa (Zen4), SMT OFF, GCC 14.2, single-threaded. ~2% run-to-run variation expected.

Level 6 (optimized)

Dataset	Baseline CS (MB/s)	Optimized CS (MB/s)	CS Speedup	Ratio Baseline	Ratio Optimized	Ratio Change	DS Change
silesia.tar	111.47	122.22	+9.6%	3.462	3.419	-1.2%	-2.6%
calgary.tar	94.31	104.77	+11.1%	3.309	3.270	-1.2%	-3.5%
canterbury.tar	111.16	124.73	+12.2%	4.708	4.650	-1.2%	-2.7%
freeBSD-13.2	421.26	443.69	+5.3%	1.610	1.608	-0.1%	-0.8%

Summary

• ~10% compression speed improvement at level 6 across all datasets
• ~1.2% ratio cost on text-heavy datasets, negligible on binary data
• ~2.5% decompression regression on text datasets, minimal on binary
• Level 5 and all other levels are completely unaffected
• Best suited for workloads where level 6 compression speed is a bottleneck and the slight ratio and decompression tradeoffs are acceptable