Zero-result refine hints and coverage truth

[catoncat]

Parent

#86

What to build

Improve zero-result behavior so agents can tell whether a miss is trustworthy. A zero-result response should distinguish fresh miss, stale/missing coverage, and over-constrained natural-language query, then suggest safe next actions without mutating the index.

Acceptance criteria

• Zero-result JSON includes a structured reason category when possible.
• Fresh coverage misses are clearly distinguishable from stale/missing coverage misses.
• Long or mixed Chinese/English queries receive deterministic broadening/refine hints.
• Read-only commands do not trigger sync or refresh implicitly.
• Tests cover fresh miss, stale coverage miss, and over-constrained query hints.

Blocked by

• Minimal recall packet in find JSON #90