feat(ai): real retrieval — Ask AI answers from the docs

[oreofeolurin]

Stacked on #11 (feat/ai-composer-textarea-attachments) — merge that first, or I'll rebase onto main once it lands.

Problem

Ask AI couldn't answer even when the answer was on the current page. Root cause: it was handed no document content — only a section-name string ("you are viewing the 'Releases' section"), and pickSources() returned hardcoded fake labels. So it replied "the provided context does not cover this." Mintlify, by contrast, retrieves the relevant page text and grounds on it.

Fix — lib/ai/retrieval.ts (client-side RAG)

1. Current page always included — reads the page's prose from the DOM (.docs-prose / .api-doc). Fixes "on the Services page it still couldn't answer."
2. Cross-page retrieval — ranks the build-time llms-full.txt corpus against the question with saturating tf·idf + title boost, plus a light stemmer (so create a service matches the Services page). Attaches the top 3 pages → answers off-page questions, Mintlify-style.
3. Real "Used N sources" — reflects the pages actually retrieved.

Pure client-side (one cached fetch of /llms-full.txt) → works in BYOK mode and on pure-static hosting; degrades to current-page-only if llms-full.txt is absent. Proxy context cap raised 8k→16k.

Validation

Ranking unit-checked against a rune-like corpus:

Question	Top result
how do i create a service	Services (1.97)
how do releases roll back on failure	Releases (3.81)
authenticate api requests	Authentication (1.93)
expose a service on a port	Services, Networking

tsc clean · next lint clean · production build prerenders all 25 pages.

Known gap

llms-full.txt is docs-only, so API-reference operation bodies aren't in the corpus yet (the current API page's DOM text is still included). Follow-up: add API operations to the corpus.