GFQL/Cypher: bounded reentry ORDER BY admits parameterized LIMIT (#1349)

[lmeyerov]

Summary

• resolve bounded reentry LIMIT values from runtime params when LIMIT $n is used in MATCH ... WITH ... ORDER BY ... LIMIT ... MATCH ... shapes
• treat resolved non-negative integer limits as bounded-safe for prefix ORDER BY reentry checks (still requires no prefix SKIP when trailing ORDER BY is absent)
• keep fail-fast behavior for unresolved/non-integer limits
• add focused pandas + cuDF regressions for parameterized LIMIT, and keep shared-GPU environments stable by using _require_cudf_runtime() in reentry cuDF tests

Issue

Part of #1349

Validation

• python -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k "parameterized_limit_shape or parameterized_non_int_limit_and_order or with_match_reentry_limit_shape"
• ./bin/ruff.sh graphistry/compute/gfql/cypher/reentry/carry.py graphistry/compute/gfql/cypher/reentry/compiletime.py graphistry/tests/compute/gfql/cypher/test_lowering.py
• ./bin/mypy.sh graphistry/compute/gfql/cypher/reentry/carry.py graphistry/compute/gfql/cypher/reentry/compiletime.py
• ssh dgx-spark (RAPIDS_VERSION=26.02, targeted cuDF tests): 1 passed, 1 skipped (shared GPU memory pressure)
• ssh dgx-spark (RAPIDS_VERSION=25.02, targeted cuDF parameterized-limit test): 1 passed