Improve benchmark realism for production scenarios

[Oaklight]

Summary

Current benchmarks focus on mean execution time against reference libraries with synthetic data. While this is a solid baseline, it misses several dimensions that matter in production: tail latency, memory usage, concurrency, data realism, and scalability curves.

This is a tracking issue for a set of improvements to make zerodep benchmarks more representative of real-world workloads.

Current strengths

• Apple-to-apple comparison against 30+ reference libraries
• S/M/L data size tiers across most modules
• Network modules (httpclient, websocket, httpserver) use real local server fixtures
• readability uses real CNN/BBC/Guardian HTML pages

Gaps vs production reality

Gap	Impact	Description
Mean-only reporting	Low	Report shows only mean; min/max/stddev already collected but not displayed
No memory metrics	Medium	Peak RSS matters in containerized deployments; parsers (yaml, soup, multipart, protobuf) are most affected
No concurrency benchmarks	High	httpclient tested async but single-request only; no asyncio.gather or thread-pool contention
Synthetic data dominance	Medium	Most modules use generated data; real configs, dirty HTML, edge-case payloads are absent
No cold start measurement	Low	First-call overhead (import, regex compile, table init) hidden by warmup rounds
Fixed S/M/L, no scale curve	Medium	Three data points can't reveal complexity inflection points

Sub-issues

• Benchmark: expose tail latency (min/max/stddev/P95) in reports #81 — Expose tail latency (min/max/stddev/P95) in reports P1
• Benchmark: add concurrency/throughput benchmarks for network modules #82 — Add concurrency/throughput benchmarks for network modules P1
• Benchmark: add real-world fixture data for parser modules #83 — Add real-world fixture data for parser modules P1
• Benchmark: add memory peak measurement for parser modules #84 — Add memory peak measurement for parser modules P2
• Benchmark: add memory peak measurement for parser modules #85 — Add scale curves for key modules P2
• Benchmark: add scale curves for key modules #86 — Measure cold start / first-call overhead P3