v0.2.0: test infra + bench harness + measured perf claims

[theYahia]

Summary

Phase A milestone per coordinated implementation plan: ship credibility (numbers + tests) before Phase B-D ops/docs work.

• Tests: 8 node:test unit tests covering cache.js public API; built-in runner, zero new devDeps
• Bench: npm run bench reports p50/p95/p99 latency for write/read hit/read miss/list, plus storage growth. JSON results saved with full machine metadata (CPU, RAM, OS, Node version, commit) for reproducible before/after comparisons. Canonical baseline committed to bench/baselines/; ephemeral runs go to gitignored bench/results/.
• CI: real npm test job replaces misnamed publish --dry-run-only workflow; publish-dry-run gated on tests green
• Docs: README replaces unmeasured "5-15x fewer WebFetch calls" headline with measured "~0.05ms cached reads"; CHANGELOG backfilled (0.1.4 unlimited TTL, 0.1.5 auto-cache hook) and 0.2.0 entry added

Bench numbers (Ryzen 3900X, Node v24.14, win32, N=10000)

Op	p50	p95	p99	ops/sec
read_hit	0.05ms	0.08ms	0.13ms	12,300
read_miss	0.01ms	0.03ms	0.05ms	69,700
write	0.06ms	0.12ms	1.30ms	8,100
list_50	0.08ms	0.13ms	0.40ms	10,600

Storage: ~1.9KB per entry on 1KB payload (overhead ~ 900B for key + indexes + WAL).

Test plan

• CI test job green on Linux/Node 22.5
• CI publish-dry-run still green (regression check)
• Local npm test 8/8 pass (Windows/Node 24.14)
• Local npm run bench produces JSON in bench/results/ + canonical baseline saved

🤖 Generated with Claude Code