Reduce repeated work in parallel visitor callbacks #296

ParallelVisitor rebuilds callback arguments and visitor indices during each node visit.
This adds overhead when many callbacks do little work.

The proposed change builds callback arguments once and stores indices with cached handlers.
It also skips action checks for None results while keeping callback order and behavior unchanged.
The full test suite passes: 3,613 passed, 18 skipped.

Measurements use CPython 3.12.13, Linux x86-64, and an AMD EPYC 9R14.
Both versions ran on the same isolated core with at least 50 rounds per case.
The base commit is f9bdf7c5c728a04bd3f27bfb8d394e9ac95751e2.
The large gain applies to empty callbacks; practical validation benefits are smaller.

Reproduction:

PYTHONPATH=src python -m pytest tests/benchmarks/test_visit.py tests/benchmarks/test_validate_gql.py tests/benchmarks/test_validate_sdl.py --benchmark-enable --benchmark-min-rounds=50

Codex helped prepare and test the proposed change, with candidates tracked through Weco Observe.
We would welcome feedback on this small Python-specific change before opening a pull request.