Add passive hydration keyword boost from FTS5

[durch]

Goal

Enhance passive hydration to include keyword-matched tool results alongside semantic results. Catches exact matches that vector search misses.

Depends on: #69 (FTS5 keyword search)

Design: .oh/designs/fts5-tool-result-search.md (Phase 2)

Context

Passive hydration currently uses LanceDB vector search only. After #69 lands, we have an FTS5 store of all tool results. This issue adds a keyword boost step to passive hydration:

1. Extract distinctive terms from the hot window (file paths, error codes, identifiers)
2. Query FTS5 with those terms
3. If any keyword results aren't already in the semantic results, append them (up to 2-3 extra entries within hydration budget)

Additive only -- doesn't replace semantic search, just catches the "exact error code" case.

Acceptance Criteria

• Passive hydration queries FTS5 after semantic search
• Keyword results that duplicate semantic results are filtered out
• Up to 2-3 keyword-only results appended within hydration budget
• Term extraction from hot window is lightweight (paths + identifiers, not full text embedding)
• Keyword boost is skippable if FTS5 store unavailable
• No measurable latency increase on the hot path (keyword query is fast)

Technical Notes

Files to modify

• packages/coding-agent/src/context/recall/passive-hydration.ts

Approach

• After searchRecall() returns semantic results, extract key terms from hot window
• Run FTS5 query via ToolResultStore.search()
• Dedup against semantic results (match by content hash or turn number)
• Append novel keyword results up to budget

[coderabbitai]

🔗 Related PRs

#48 - Ingest pipeline: persist and embed all messages into LanceDB [merged]
#49 - Passive hydration: auto-inject relevant past context each turn [merged]
#50 - Recall tool: active session history search for the agent [merged]
#57 - Global recall store for cross-project search [merged]
#61 - Add source provenance tags to assembled context [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!