Fixes #1411

Summary

• add native GFQL row-pipeline join_apply(binding_ops, join_aliases, how) for joining active binding rows to correlated binding rows
• register/safelist the new row-pipeline call
• add direct GFQL regression coverage for inner, left, and cuDF joined row projection, plus Cypher guard coverage for already-working connected joins
• harden DGX RAPIDS 25.02 image builds by injecting the CUDA Python bridge version that avoids cuDF host-transfer segfaults on DGX Spark GB10

Validation

• PYTHONPATH=/tmp/pyg1411-deps python3 -m pytest -q graphistry/tests/test_compute_chain.py::TestChainBindingsTable::test_issue_1411_join_apply_projects_joined_message_rows graphistry/tests/test_compute_chain.py::TestChainBindingsTable::test_issue_1411_join_apply_projects_joined_message_rows_on_cudf graphistry/tests/test_compute_chain.py::TestChainBindingsTable::test_issue_1411_join_apply_left_preserves_unmatched_rows graphistry/tests/compute/gfql/cypher/test_lowering.py::test_issue_1411_connected_join_property_projection_shape graphistry/tests/compute/gfql/cypher/test_lowering.py::test_issue_1411_connected_join_whole_row_projection_shape -> 4 passed, 1 skipped locally
• PYTHONPATH=/tmp/pyg1411-deps PATH=/tmp/pyg1411-bin:$PATH ./bin/test-minimal-lite.sh --ignore=plans --ignore=test_env --ignore=graphistry/tests/test_text_utils.py --ignore=graphistry/tests/test_compute_cluster.py --ignore=graphistry/tests/compute/test_indegrees_after_umap.py --ignore=graphistry/tests/compute/test_let_engine_coercion.py --ignore=graphistry/tests/compute/test_safe_map_series.py --ignore=graphistry/tests/compute/test_call_engine_coercion.py -k 'not cudf and not umap and not dask' -> 2947 passed, 155 skipped, 244 deselected, 15 xfailed
• ./bin/ruff.sh graphistry/tests/test_compute_chain.py graphistry/compute/ast.py graphistry/compute/gfql/call/validation.py graphistry/compute/gfql/row/pipeline.py graphistry/tests/compute/gfql/cypher/test_lowering.py
• bash -n docker/test-rapids-official-local.sh
• git diff --check
• DGX RAPIDS 26.02: streamed local worktree to ssh dgx-spark, ran docker/test-rapids-official-local.sh with RAPIDS_VERSION=26.02 PROFILE=gfql WITH_GPU=1 WITH_IMAGE_BUILD=0, targeted [POST-RERUN] #880 residual: joined row projection bucket #1411 cuDF selector -> 1 passed
• DGX RAPIDS 25.02: rebuilt graphistry/test-rapids-official:25.02-gfql with PIP_PRE_DEPS=numba-cuda==0.22.2 cuda-bindings==12.9.5 cuda-core==0.3.2 cuda-python==12.9.5, targeted [POST-RERUN] #880 residual: joined row projection bucket #1411 cuDF selector -> 1 passed, 1 warning
• DGX RAPIDS 25.02 smoke: rebuilt image now passes independent cudf.DataFrame({"x": [1, 2]}).to_pandas() host-transfer smoke; package bridge was {'numba_cuda': '0.22.2', 'cuda_bindings': '12.9.5', 'cuda_python': '12.9.5'}
• DGX RAPIDS 25.02 full GFQL profile: rebuilt current worktree with WITH_IMAGE_BUILD=1 RAPIDS_VERSION=25.02 PROFILE=gfql; default GFQL suite (test_parser.py, test_row_pipeline_ops.py, and cuDF lowering selectors) -> 343 passed, 6 warnings
• DGX RAPIDS 26.02 full GFQL profile: rebuilt current worktree with WITH_IMAGE_BUILD=1 RAPIDS_VERSION=26.02 PROFILE=gfql; default GFQL suite (test_parser.py, test_row_pipeline_ops.py, and cuDF lowering selectors) -> 343 passed, 5 warnings

Known validation caveats

• local ./bin/typecheck.sh on touched runtime files currently reports graphistry/_version.py:57 HANDLERS annotation, outside this diff
• local ./bin/test-minimal-lite.sh hardcodes python; this shell only has python3, so validation used an explicit /tmp/pyg1411-bin/python -> /usr/bin/python3 shim
• local host has cuDF installed but no CUDA-capable device, so the cuDF-specific [POST-RERUN] #880 residual: joined row projection bucket #1411 smoke test skips locally
• rebuilt DGX RAPIDS 25.02 emits a cudf _ptxcompiler.py warning about cuDriverGetVersion() / Not patching Numba; import smoke, trivial cuDF host transfer, and the targeted [POST-RERUN] #880 residual: joined row projection bucket #1411 cuDF selector pass