perf(accesslog): add sample_rate for access log to reduce CPU and allocations
Add configurable access log sampling via :
- 0.0-1.0 range; defaults to 1.0 (record all) for backward compatibility
- Uses lock-free atomic counter for deterministic sampling
- Non-2xx responses always logged regardless of sample rate
Benchmark results (combined format, /dev/null):
Full logging: ~2245 ns/op, 1987 B/op, 17 allocs/op
10% sampling: ~1593 ns/op, 1633 B/op, 6 allocs/op
Improvement: -29% latency, -65% allocations/op
This addresses the top application-layer CPU hotspot identified
in the v0.4.0 profile (LogAccess at 16.36% cumulative CPU).
xfy
445401c40f