fix(cypher): disambiguate IC4 HAS_TAG endpoints

[lmeyerov]

Closes #1496.

Summary

• Disambiguate unlabeled forward HAS_<Label> destination bindings when graph node ids collide across labels and the implied label__<Label> column exists.
• Add an IC4/new-topics regression using the official CASE aggregation shape with LDBC-style numeric id collisions between Post and Tag rows.
• Update CHANGELOG.md.

Why

pyg-bench#4 showed direct-Cypher IC4 executing but returning the wrong tag set after the CASE blocker was fixed. The in-repo IC4 tests used globally unique ids, while the benchmark graph can have cross-label numeric id collisions. In that shape, -[:HAS_TAG]->(tag) is unlabeled, and the connected row-binding join could bind the colliding Post row as tag before sum(valid) / sum(inValid) and the post-aggregation WHERE.

Validation

• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py::test_issue_1496_ic4_unlabeled_has_tag_destination_disambiguates_colliding_ids[pandas]
• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k 'issue_1496 or ic4 or 1413'
• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py -k '(multi_alias or issue_1496 or issue_1413 or binding_rows or rows) and not cudf'
• ./bin/ruff.sh graphistry/compute/gfql/row/pipeline.py graphistry/tests/compute/gfql/cypher/test_lowering.py
• ./bin/typecheck.sh graphistry/compute/gfql/row/pipeline.py
• git diff --check
• DGX RAPIDS 25.02 focused cuDF regression: passed
• DGX RAPIDS 26.02 focused cuDF regression: passed

Note: the local broader slice without and not cudf hit unrelated cudaErrorNoDevice failures because this workstation imports cuDF but has no CUDA device. cuDF validation was done on dgx-spark.

pyg-bench follow-up

After this lands, rerun graphistry/pyg-bench#4 IC4/new-topics direct-Cypher conformance against a prepared pyg-bench checkout with the SF0.1 fixture:

uv run python scripts/run_dgx_spark_suite.py \
  --suite snb-interactive \
  --config configs/suites/snb-interactive-ic4-conformance-sf0.1.yaml \
  --output-dir results/runs/dgx-spark-snb-interactive-ic4-pygraphistry-<merge-sha>-sf0.1-r1 \
  --graphistry-repo-path /path/to/pygraphistry/graphistry

If SF0.1 reports ok, then consider the SF1/no-warmup rerun only if coordinator wants a scale confirmation; prior notes say SF1 can OOM on the shared DGX box.