Add semantic search to the UI and surface similar moments for bookmarks

[chigichan24]

Why

• The Overview / Knowledge Graph views let users navigate by topic, keyword, or facets. Natural-language queries like "places where I configured Vite" or "auth-related debugging" cannot be answered today.
• Once bookmarks (Bookmark / tag / annotate valuable moments in Session Playback (human feedback channel) #23) exist, the natural follow-up is "show me other moments like this one" — the strongest UX for human-in-the-loop tagging and skill curation.
• This also addresses one of the questions raised in [Discovery] Rethink the Knowledge Graph: which questions is it actually answering? #19 ("which past sessions are most similar to my current work") more directly than graph navigation can.

What

• Add a search bar to Overview / Knowledge Graph that calls into the retriever and lists matching turns with their session context.
• Add a "Similar to this bookmark" affordance on each bookmarked turn.
• Optional: cluster / community-aware diversification so results are not all from one session.

How

• skill-server exposes a new endpoint /api/retrieve that wraps the retriever (Add embedding pipeline and chunk-level vector index for RAG-based retrieval #32).
• Frontend: a small client hook (useSemanticSearch(query)) hitting that endpoint and rendering results that link into existing playback navigation.
• For purely-static deployments, in-browser retrieval is workable if the index size after quantization stays modest; decide based on PoC measurements.

Depends on

• Add embedding pipeline and chunk-level vector index for RAG-based retrieval #32 — embedding pipeline + retriever
• Bookmark / tag / annotate valuable moments in Session Playback (human feedback channel) #23 — bookmarks (only for the bookmark-similarity surface; the search bar is independent)

Related

• [Discovery] Rethink the Knowledge Graph: which questions is it actually answering? #19 — Knowledge Graph rethink. The search-bar UX may directly answer some of the questions originally aimed at the graph; worth comparing both side by side.