feat(docs): publish per-model concurrency with tier numbers and per-key ceiling

[sre-helmcode]

Closes #79

Summary

• rateLimits.ts: per-model concurrency table (tierMaxParallel), per-key outer ceiling line, exempt-endpoints note, corrected glm5.2 comment (hidden by owner decision, served via alias).
• RateLimits.astro per-key + per-model cards (EN/ES); pricing premium bullet 5 -> 10; mdxToText + OpenAPI spec kept in sync; deterministic fixture.
• All pinning tests updated; no flat per-key '5 concurrent' claim remains.

Test evidence

• npm run build: Complete. npm test: 56 files / 1162 tests passed.