Route simple OPTIONAL MATCH through native physical dispatch

[lmeyerov]

Summary

Fixes #1460.

Simple input-free top-level OPTIONAL MATCH now gets a native logical plan and dispatches through the same-path physical route instead of _execute_compiled_query_compat_non_union().

Scope

• Plan first-clause top-level OPTIONAL MATCH as PatternMatch(optional=True, arm_id="top_level_optional_0").
• Keep non-top-level/reentry optional shapes deferred for the follow-up deletion lane.
• Add runtime cutover tests proving matched and unmatched/null-extension top-level optional queries bypass the compat executor.
• Update changelog.

Validation

• python3 -m pytest -q graphistry/tests/compute/gfql/test_logical_planner.py graphistry/tests/compute/gfql/test_runtime_physical_cutover.py graphistry/tests/compute/gfql/cypher/test_lowering.py::test_compiled_query_sets_logical_plan_route_for_top_level_optional_shape graphistry/tests/compute/gfql/cypher/test_lowering.py::test_compiled_query_sets_logical_plan_defer_reason_for_optional_reentry_shape -> 37 passed
• python3 -m pytest -q graphistry/tests/compute/gfql/cypher/test_lowering.py graphistry/tests/compute/gfql/test_runtime_physical_cutover.py graphistry/tests/compute/gfql/test_logical_planner.py graphistry/tests/compute/gfql/test_physical_planner.py -k 'not cudf' -> 976 passed, 12 skipped, 111 deselected
• ./bin/ruff.sh graphistry/compute/gfql/logical_planner.py graphistry/compute/gfql/cypher/lowering.py graphistry/tests/compute/gfql/test_logical_planner.py graphistry/tests/compute/gfql/cypher/test_lowering.py graphistry/tests/compute/gfql/test_runtime_physical_cutover.py -> green
• ./bin/typecheck.sh -> green
• git diff --check -> clean
• DGX RAPIDS 25.02: RAPIDS_VERSION=25.02 PROFILE=gfql WITH_GPU=1 WITH_IMAGE_BUILD=0 TEST_FILES="graphistry/tests/compute/gfql graphistry/tests/compute/test_gfql.py" docker/test-rapids-official-local.sh -> 2368 passed, 22 skipped, 15 xfailed
• DGX RAPIDS 26.02: RAPIDS_VERSION=26.02 PROFILE=gfql WITH_GPU=1 WITH_IMAGE_BUILD=1 TEST_FILES="graphistry/tests/compute/gfql graphistry/tests/compute/test_gfql.py" docker/test-rapids-official-local.sh -> 2368 passed, 22 skipped, 15 xfailed
• Review skill loop: 2 waves, credentials gate clean, no carried findings

Local cuDF/GPU suites fail on this workstation with cudaErrorNoDevice; DGX GPU validation covered both requested RAPIDS versions.