v0.9.6: backend load-test harness (warm /analyze)

[Shaan-alpha]

Summary

A reusable open-loop load-test harness for the backend warm /analyze path. The harness is the durable deliverable; the full 100 RPS validation run is an operator step (hardware-gated).

• backend/loadtest/run.py — open-loop (fixed-rate) async dispatcher (httpx, already a dep), p50/p95/p99 + error rate + achieved RPS, pass/fail thresholds, ramp, exit 0/non-zero.
• 6 unit tests for the pure stats helpers (backend/tests/loadtest/test_stats.py).
• backend/loadtest/README.md runbook — local SRH (Docker) warm-cache setup, prime-then-measure, ramp, point-at-deploy, Windows/Git-Bash MSYS_NO_PATHCONV note.
• No application-code change (relies on the existing "anonymous + unset INTERNAL_PROXY_SECRET → analyze limiter skips enforcement" behavior).

Test plan

• Backend ruff clean; pytest 290 passed / 69 skipped (+6 stats tests)
• Frontend lint/tsc clean; vitest 54; next build clean (version literals only)
• Harness sanity run (backend-only): /openapi.json 10 RPS × 5 s → 0 errors, p95 6.2 ms, PASS, exit 0
• Full 100 RPS warm-/analyze run — operator step (Docker/SRH + GITHUB_TOKEN), recorded when run