fix(gfql): polars declines prune_to_endpoints on single-hop edges (#2053)

[lmeyerov]

Stacked on #2062. Closes #2053.

Defect. The polars chain implements prune_to_endpoints for variable-length hops through its auto hop labels, but a single-hop edge carrying the flag was admitted by the plain single-hop branch and by the full polars traversal and silently returned the seed and the arrival side where pandas keeps only the arrival side. Polars and pandas disagreed on every single-hop prune shape.

Fix. A single-hop edge with prune_to_endpoints raises the polars engine's typed NotImplementedError (use engine='pandas' or 'cudf'), and polars_plain_single_hop_admits declines the shape. Variable-length hops are untouched (their native pruning has pins in test_row_pipeline_boundaries.py / test_path_trail_semantics.py, which caught the first, too-wide cut). Typed decline over silent divergence; the single-hop arrival-side contract on polars is a follow-up once the plain branch carries hop labels.

Pins (tests/compute/gfql/lazy/engine/polars/chain_specializations/test_polars_admission.py): seeded and unseeded single hop raise on polars while pandas answers [2, 3]; the predicate declines the shape and still admits the plain seeded hop. The harness KNOWN entry and the corpus / _BYPASS_SHAPES #2053 tags are gone (the cells are no longer admitted on polars-plain).

Local at this head: row-pipeline boundaries + trail semantics + polars chain_specializations + routes 226 pass / 11 xfail; guards + mypy green; broad suites rerunning, posted below.

🤖 Generated with Claude Code

https://claude.ai/code/session_01QztW7jYsDd66e8rb8pJNQA

[lmeyerov]

Local receipt at 198cbd5 (TEST_CUDF=1, cuDF 25.10 local): test_chain/test_hop/test_gfql + tests/compute/gfql + chain_specializations + collision + closure-dtype suites 12839 passed / 0 failed / 37 xfailed. The first cut (decline on every hop shape) failed two variable-length pins on polars (test_row_pipeline_boundaries, test_path_trail_semantics), which is how the scope narrowed to single-hop edges. CI pending; the dgx cuDF + polars-gpu lane for this head follows the stack's next batch.

[lmeyerov]

CI receipt at 198cbd5: 67 check-runs success, 7 path-skipped as whole jobs (tck-gfql, test-polars ×6, test-gfql-core, gfql-benchmarks and the guards green). Local broad suites 12839 / 0 failed. Marking ready; the dgx lane for this head rides the next stack batch (the change is a polars-only typed decline; the cuDF lane is unaffected).

[lmeyerov]

GPU receipt (dgx-spark, graphistry/test-rapids-official:26.02-gfql-polars, cudf 26.02.01 / polars 1.35.2, safe_run, TEST_CUDF=1; suites test_chain.py, test_hop.py, test_chain_alias_column_collision.py, tests/compute/gfql/**) on the restacked stack head a54ac4b (#2065, containing #2055 rework, #2056, #2054, #2061, #2062): 13386 passed / 95 failed / 38 xfailed / 4 xpassed. 93 of the failures are master 86de0f5's polars-gpu fused-lane set (#2064, fixed separately in #2066). The other 2 are strict XPASS of the #2043 prune_to_endpoints markers on cuDF 26.02: the prune shapes now agree fast-vs-full on this stack (the #2062 closure fix removed the divergence), so #2062 drops that marker in a follow-up commit (tree otherwise identical; #2065 restacked).

[lmeyerov]

CI receipt at 730aadf (restacked): 67 check-runs success, 7 path-skipped as whole jobs. GPU receipt on the stack head above.