GFQL Cypher: run the TCK query corpus through the LALR≡Earley parser differential (parser axis)

[lmeyerov]

Follow-up from the Earley→LALR parser PR (#1682).

Context

tck-gfql conformance already runs green on the new LALR parser (semantic axis — queries produce openCypher-correct results). The parser PR also ships a parser-axis differential harness (test_differential_grammar_parity.py) that runs every query through the production pipeline under both LALR and Earley and asserts byte-identical ASTs + rejection parity.

Ask

Feed the TCK query corpus through that same LALR≡Earley differential harness in tck-gfql CI, to get parser-equivalence coverage over the full TCK surface (not just the repo-scraped corpus). The harness is corpus-parameterizable — this is a tck-gfql-repo wiring change, no pygraphistry parser work needed.

🤖 tracked via Claude Code

[lmeyerov]

Refinement (from review): rather than always running the TCK parser-axis differential, gate it on parser changes via conditional CI steps — it only needs to run when the grammar/parser actually changed.

• In pygraphistry CI this maps to the existing cypher_frontend_ci path filter (graphistry/compute/gfql/cypher/, .../ir/, .../frontends/cypher/), so the differential step runs only on parser-touching PRs.
• If the harness lives in the tck-gfql repo, mirror the same idea there: a paths:-filtered job (or a step if: gated on a parser-path change) so routine TCK/data-only changes don't pay for it.

This keeps the parser-equivalence coverage without adding steady-state CI cost. Related: the broader CI over-triggering audit in #1689 (today the python catch-all filter would run this regardless — the fix there makes narrow gating like this actually effective).