GFQL: admit bounded ORDER/LIMIT MATCH reentry top-k (#1342)

[lmeyerov]

Summary

• relax bounded MATCH-reentry prefix ordering guard to admit WITH+ORDER BY+LIMIT literal for multi-row top-k prefixes
• keep conservative guards for unbounded ORDER BY and parameterized LIMIT
• add lowering regression coverage for ordered top-k reentry (single-column + multi-column) plus cuDF parity test
• update CHANGELOG.md development section

Why
Issue #1342 (under #880 umbrella) tracks a benchmark-visible gap where ordered top-k prefixes were rejected before trailing MATCH re-entry, even though runtime reentry seeding already preserves prefix row order.

Validation

• python -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k with_match_reentry_limit_shape or ordered_topk_multi_row_shape or ordered_topk_multicolumn_shape or rejects_reentry_with_parameterized_limit_and_order or failfast_rejects_with_match_reentry_unsupported_shapes
• python -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k with_match_reentry and (order or limit)
• uv run ruff check graphistry/compute/gfql/cypher/reentry/carry.py graphistry/tests/compute/gfql/cypher/test_lowering.py

Issue links

• partial for [FEA] GFQL bindings_table (row bindings) from chain using Yannakakis prune #880
• addresses Cypher WITH ORDER BY / LIMIT not preserved across MATCH re-entry for multi-row results #1342