[lmeyerov]

Generic core CI reruns the full-repository LALR/Earley grammar differential test in seven Python cells. This PR excludes only that file from the generic core matrix; test-gfql-core still runs it on Python 3.12 (with coverage) and 3.14, and the dedicated differential gate and developer bin/test.sh default stay unchanged. The core matrix still runs the constructed grammar invariant tests on every supported Python version.

In #2181's baseline hosted core logs, the duplicate occupied 285.4/273.7/251.0/86.9/96.6/114.6/83.4 seconds on Python 3.8–3.14 respectively: 19.86 runner-minutes of directly identified repeated work. On this PR's exact f86d2766d head, all seven core cells passed and totaled 32.58 runner-minutes, versus 40.85 on recent master and 41.23 on #2181. That observed 20–21% core-group reduction includes runner variance; it is not a precise causal speedup estimate. The full workflow took 17m46s, but differing path-triggered jobs and host conditions prevent an apples-to-apples wall-time claim.

The GFQL core trigger subsumes the generic core trigger across all 2,048 checked path-filter/event states. Exact-head hosted CI completed with 79 successful checks and two path-gated skips. Logs confirm the sweep is absent from core 3.8/3.14 and passes in GFQL 3.12/3.14; the dedicated parity job passed. GFQL's 3.12 per-file audit reports cypher/parser.py at 90.29% against an 88.35% baseline. The combined changed-line gate passed, but this workflow-only diff has zero eligible changed package lines, so that gate alone is not parser-coverage evidence. The full-repository sweep no longer runs on generic core Python 3.8–3.11/3.13; that reduced version scope is intentional and explicit.

This is separate from #2181, whose reviewed route-replay head is unchanged. If #2181 merges first, rerun CI on the integrated head. Pygraphistry merge remains with the maintainer.

Related: #2182.