Extract reentry helper ownership from lowering

[lmeyerov]

Summary

Closes #1497.

• Moved bounded-reentry helper ownership out of graphistry.compute.gfql.cypher.lowering into graphistry.compute.gfql.cypher.reentry.lowering_support.
• Updated reentry compile-time/rewrite imports to use the focused helper owner instead of private helpers on the monolithic lowering module.
• Kept behavior unchanged; this is a pure ownership/DRY refactor plus a split-guard test update.
• Updated CHANGELOG.md.

Validation

• python3 -m compileall -q graphistry/compute/gfql/cypher/lowering.py graphistry/compute/gfql/cypher/reentry
• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering_s3_split_guard.py graphistry/tests/compute/gfql/cypher/test_lowering.py -k "(reentry or bounded or carry or alias) and not cudf" -> 277 passed, 804 deselected
• ./bin/ruff.sh graphistry/compute/gfql/cypher/lowering.py graphistry/compute/gfql/cypher/reentry graphistry/tests/compute/gfql/cypher/test_lowering.py graphistry/tests/compute/gfql/cypher/test_lowering_s3_split_guard.py
• ./bin/typecheck.sh -> Success: no issues found in 264 source files
• python3 bin/ci_cypher_surface_guard.py
• git diff --check
• Naming follow-up: renamed broad reentry.helpers module to reentry.lowering_support; repeated compileall, focused pytest, ruff, typecheck, surface guard, and diff check locally.
• gh pr checks 1506 --repo graphistry/pygraphistry --watch -> green again on d28d6606 after the naming-only rename; includes tck-gfql, test-docs, and test-full-ai.
• DGX RAPIDS 25.02 full GFQL on rebased head c0206b8e: 2404 passed, 22 skipped, 15 xfailed (/tmp/pygraphistry-1497-dgx/rapids-25.02-gfql-c0206b8e.log); follow-up d28d6606 is naming-only and locally revalidated.
• DGX RAPIDS 26.02 full GFQL on rebased head c0206b8e: 2404 passed, 22 skipped, 15 xfailed (/tmp/pygraphistry-1497-dgx/rapids-26.02-gfql-c0206b8e.log); follow-up d28d6606 is naming-only and locally revalidated.

Local cuDF-focused run was attempted first, but this shell has cudf installed without a CUDA device, so cuDF tests fail at cudaErrorNoDevice before exercising this change. DGX 25.02/26.02 full GFQL is green.

Review

• Review skill converged: wave 1 fixed one stale docstring; waves 2 and 3 had no findings/no advance; wave 4 handled the naming review with no behavioral findings.

Coordination

• Rebased cleanly over current origin/master after fix(cypher): disambiguate IC4 HAS_TAG endpoints #1502 and chore(gfql): remove unused cypher lowering helper #1500 landed.
• Do not auto-merge.