Feat/redirect latency baseline

[Yentec]

Description

Baseline redirect latency measurement using k6

Changes

• Added tests/load/redirect.js — k6 script (100 VUs, ramp 0→100→0 over 60s), redirects: 0 to measure only the server response, summaryTrendStats including p(99), informational thresholds not enforced in CI
• Added tests/load/ to ESLint ignores — k6 script is outside the TypeScript project scope; __ENV is a k6 global not recognized by the TypeScript parser
• Made rate limit maximum configurable via RATE_LIMIT_MAX environment variable (default: 100) — in src/config/env.ts + src/app.ts; required for local benchmarking without being capped at 100 req/min per IP
• Updated .env.example with RATE_LIMIT_MAX (commented, documented benchmark value)
• Added docs/adr/0007-redirect-latency-baseline.md — documents methodology, measured results, and confirms ADR 0005 assumption (async tracking → latency bounded by cache performance)
• Updated README.md — added Performance section with reproducible Docker benchmark commands and metrics table; added ADR 0006 + ADR 0007 to Decisions; removed "Refresh-token-reuse detection" from roadmap (already implemented in PR Feat/refresh token reuse detection #15)

Validation

• Benchmark: 91,936 requests, 0 failures, p50=45 ms / p95=88 ms / p99=103 ms / ~1,530 req/s
• Clean typecheck
• Clean lint
• No regression in existing tests (RATE_LIMIT_MAX default=100, unchanged behavior in tests)
• RATE_LIMIT_MAX=1000000 locally → rate limiting effectively disabled for benchmarking → 0% failures

Closes #11