Solution for gh-ppx_minidebug-84

[lukstafi]

Solution from coder for feature: gh-ppx_minidebug-84

[lukstafi]

Refactoring Suggestions

Post-merge retrospective: what would we do differently if starting from scratch?

Refactoring Suggestions

If starting from scratch, I would...

• Factor duplicated entry decoding into a single helper in client/query.ml (e.g., entry_of_stmt) used by search_entries, find_entry, find_scope_header, streaming search, and extract-search paths. This would remove repeated Sqlite3.column mapping blocks and reduce regression risk when entry schema handling changes.
• Introduce a shared DAG-aware propagation helper in client/query.ml used by both populate_search_results and populate_extract_search_results (including direct-parent insertion + ancestor traversal + quiet-path stop semantics). The two code paths currently duplicate subtle propagation behavior and can drift.
• Add a public query API for "all headers for child scope" in client/query.mli/client/query.ml (find_scope_headers), and migrate internal callers from ad-hoc SQL fragments. This would make DAG semantics explicit instead of relying on hidden local helpers.
• Split populate_extract_search_results in client/query.ml into smaller units: candidate enumeration, dedup bookkeeping, highlight insertion, and propagation. The current monolithic function mixes concerns, making edge-case reasoning difficult.
• Add a deterministic regression test fixture utility in test/ for resolving latest run DB via metadata (*_meta.db), so tests do not hardcode _1.db and do not depend on filesystem cleanup state.
• Add one focused test that validates quiet-path stopping behavior in DAG propagation for both search and extract modes, to lock down the intended semantics when a duplicated ancestor matches quiet-path.