Add cqp_query tool

Armaan Guptarequested to merge
tools/cqp-query into develop

Closes #12 (closed)

Description:

Adds cqp_query, the corpus-search tool. The model writes a CQP query; we run it against the cqp4rdf backend and return the matches as KWIC lines.

What it does

  • Runs CQP queries against cqp4rdf.

  • Returns up to 50 KWIC lines per page, e.g.:

    P100065 1(u) gin2 ku3-babbar … 2(u) sze gur [lugal] i3-ag2-e-a mu …

  • Each line starts with the source tablet's P-number, so the model can hand it to get_metadata / get_inscription.

  • Endpoint is configurable via CQP_BASE_URL (defaults to the deployed CDLI endpoint).

Handling the backend's quirks

  • Multi-word queries must include a :: join, without one the backend computes a Cartesian product and always times out. The tool blocks un-joined multi-word queries up front with a clear message.
  • Even with a join, the backend currently times out on most multi-word queries. This is a known backend limitation (documented in the tool description); single-word queries are reliable.
  • The three outcomes are worded so the model can't confuse them:
    • no matches → "query ran fine, nothing matched, don't retry"
    • timeout → "query was too expensive, matches may exist, narrow it and retry"
    • bad syntax → upstream error, revise the query

Merge request reports