test(cypher): harden two-MATCH reentry varlen regression coverage (#1001)

[lmeyerov]

Summary

Hardens coverage around #1001 (two-MATCH reentry + varlen wrong-row concern) by moving from permissive shape checks to exact-result and equivalence assertions.

What changed

• Strengthened existing acceptance test to assert concrete expected rows instead of only result._nodes is not None
• Added forward regression: split reentry query equals connected single-MATCH query on a non-empty deep-tree fixture
• Added reverse regression: split reentry query equals connected single-MATCH query on a non-empty reverse-direction fixture

Why

Issue #1001 reported wrong rows for TCK-style queries match5-25/26. Current master behavior was not reproducible as a product wrong-row bug on valid standalone fixtures; this PR locks that behavior with explicit tests to prevent regressions.

Validation

• PYTHONPATH=. UV_CACHE_DIR=/tmp/uv-cache uv run --no-project --with pytest --with pandas python -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k "nonterminal_variable_length_relationship_patterns or two_match_reentry_varlen"
• PYTHONPATH=. UV_CACHE_DIR=/tmp/uv-cache uv run --no-project --with pytest --with pandas python -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k "reentry"
• ./bin/ruff.sh graphistry/tests/compute/gfql/cypher/test_lowering.py

Refs #1001