Issue: mergeHybridResults ignores textScore when keyword and vector results have non-overlapping chunk IDs

Summary

In hybrid memory search, mergeHybridResults only updates textScore when keyword and vector results share the same chunk ID. When they don't overlap (common with CJK/trigram queries), the keyword search results are effectively discarded — textScore stays 0 even though searchKeyword returned valid results.

Environment

• OpenClaw version: 2026.6.5
• FTS tokenizer: trigram
• Embedding provider: Ollama (qwen3-embedding:4b)

Root Cause

In extensions/memory-core/src/host/search/manager.ts, mergeHybridResults merges keyword and vector results by chunk ID:

for (const r of params.vector) byId.set(r.id, {
    ...r,
    vectorScore: r.score,
    textScore: 0  // Initialize to 0
});
for (const r of params.keyword) {
    const existing = byId.get(r.id);
    if (existing) {
        existing.textScore = r.textScore;  // Only updates if ID matches!
    } else {
        byId.set(r.id, { ...r, vectorScore: 0, textScore: r.textScore });
    }
}

When keyword results have different chunk IDs than vector results:

1. Vector results keep textScore: 0
2. Keyword-only results are added with vectorScore: 0
3. Final score = vectorWeight * 0 + textWeight * textScore → very low, often below minScore

Reproduction

1. Configure agents.defaults.memorySearch.store.fts.tokenizer: "trigram"
2. Index memory files containing Chinese text
3. Run memory_search with a multi-character CJK query (e.g., "飞书插件配置")
4. Observe: searchKeyword returns rows with usedMatch: true, but final results show textScore: 0

Logs confirm keyword search works:

[DEBUG] searchKeyword rows: 2 usedMatch: true matchQuery: "飞书插件配置" substringTerms: 6

But the final output:

{ "textScore": 0, "vectorScore": 0.597 }

Expected Behavior

Keyword search results should contribute to textScore even when chunk IDs don't perfectly overlap with vector results. Possible approaches:

1. Union merge: Add keyword-only results as separate entries (already partially done, but they get filtered by minScore)
2. Relaxed matching: Match on path + startLine range instead of exact chunk ID
3. Score blending: When no ID overlap, blend keyword score into textScore based on content similarity

Impact

• CJK queries with trigram tokenizer always have textScore: 0 in hybrid mode
• English queries may also be affected when FTS5 and vector indexes produce different chunk boundaries
• Vector search still works (vectorScore is correct), but the BM25 component contributes nothing to the final score