Summary

• add bounded vectorized Cypher reentry support for the same-alias MATCH ... WITH ... MATCH ... subset
• carry identifier-style scalar projections through the bounded reentry path for trailing RETURN / ORDER BY
• harden the compiler/runtime contract around seed ordering, null-id alignment, renamed aliases, and hidden-column collisions
• keep larger cleanup and redesign work out of scope for this PR; track them separately in #987 and #989

Supported boundary

• supported:
  • safe whole-row reentry on the bounded subset
  • carried scalar projection/order on the carried alias
• explicit failfast:
  • cross-alias carried filters
  • fresh row-seeded reentry
  • shapes that rely on prefix WITH row ordering surviving into a multi-row trailing MATCH

Validation

• local static checks:
  • python -m py_compile graphistry/compute/gfql_unified.py graphistry/compute/gfql/cypher/lowering.py graphistry/tests/compute/test_gfql.py graphistry/tests/compute/gfql/cypher/test_lowering.py
  • ruff check --no-cache graphistry/compute/gfql_unified.py graphistry/compute/gfql/cypher/lowering.py graphistry/tests/compute/test_gfql.py graphistry/tests/compute/gfql/cypher/test_lowering.py
• local targeted tests:
  • PYTHONPATH=. pytest -q graphistry/tests/compute/test_gfql.py -k 'reentry'
  • PYTHONPATH=. pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k 'reentry'
• local broad touched-file sanity:
  • PYTHONPATH=. pytest -q graphistry/tests/compute/test_gfql.py graphistry/tests/compute/gfql/cypher/test_lowering.py
  • result: 487 passed, 47 skipped
• hosted CPU CI:
  • gh pr checks 975
  • result: fully green on head b4f730fbd
• targeted GPU validation:
  • ssh dgx-spark shared-checkout remote worktree
  • official RAPIDS 26.02 cuda13
  • bounded-reentry cudf slice result: 7 passed in 3.18s

Companion / follow-ons

• companion sibling bookkeeping PR:
  • graphistry/tck-gfql#12 test: refresh direct cypher support after rebase
• architecture follow-ons kept out of this PR:
  • #987 explicit ReentryPlan / hidden-column-handshake cleanup
  • #989 broader row-carrier IR direction

Notes

• this PR intentionally stays narrower than a generic WITH or row-carrier redesign
• the remaining significant cleanup is architectural, not required to safely land the bounded-reentry feature in this branch