fix(gfql): apply single-alias predicate pushdown masks by position (#2020)

[lmeyerov]

MATCH (a)-[e]->(t) WHERE a.type IN [...] AND e.e_type IN [...] AND t.type IN [...] RETURN t.id raised GFQLTypeError [invalid-node-reference] ... Unalignable boolean Series provided as indexer on pandas (and cuDF) while polars answered. The single-alias predicate pushdown filtered the alias frame with frame.loc[mask]; once an earlier pushdown had narrowed that frame, its labels were no longer a RangeIndex, while the mask evaluated on the renamed view came back with a fresh positional index, so label alignment failed. A mask computed on the same rows is positional by contract, so both pushdown sites (expression and searchAny) now keep rows by position through one helper typed with the repo's engine-agnostic aliases.

Pins (test_alias_prefilter_alignment_2020.py): the issue's query returns ['tx1'] on pandas, polars, cuDF; equals the scalar = form; and survives a node frame whose index is neither ranged nor ordered. Row-pipeline, lowering, alias-scoping, optional-match and multiplicity suites pass locally (1885; the two cuDF t6 lowering failures are the dev-box libcublas environment, identical on master). Real-GPU run to follow on dgx.

Fixes #2020.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QztW7jYsDd66e8rb8pJNQA

[lmeyerov]

Real-GPU receipt (dgx-spark, RAPIDS 26.02 image, cudf 26.02.01): test_alias_prefilter_alignment_2020.py + tests/compute/gfql/row + polars alias-prefilter + cypher/test_lowering.py at 06e1f48 = 1689 passed, 0 failed; the 5 skips are igraph-only lowering tests (this image has no python-igraph), not GPU skips.

[lmeyerov]

READY (84ce9d0): 77/77; positional pushdown mask (#2020) with pins on pandas/polars/cuDF; real-GPU 1689 passed (igraph-only skips); SeriesT/DataFrameT typed, no new ignore.