perf: reduce benchmark suite runtime from 8min to 35s

[bonega]

Summary

• Add size-based timing configuration to reduce benchmark overhead for small inputs
• Reduce nresamples from 100k to 10k for faster bootstrap analysis
• Achieves ~93% reduction in benchmark runtime while maintaining statistical validity

Changes

Input Size	Warmup	Measurement
≤64 bytes	100ms	200ms
65-512 bytes	100ms	300ms
>512 bytes	200ms	500ms

Variance Testing (3 runs)

Category	Avg CV%
ASCII fast path	0.7%
Small sizes (8-64)	2.3%
Medium sizes (256-512)	3.8%
Large sizes (1024+)	3.2%
Total runtime	0.3%

Test plan

• Run benchmarks 3 times to verify consistent ~35s runtime
• Verify CV% remains acceptable (<5% for most benchmarks)