Handle zero Jacobian partial derivatives

[yuelgrace1810-ops]

What changed

• ignore explicit zero coefficients emitted by SymPy while assembling graded Jacobian coefficient maps
• add an end-to-end compute-and-verify regression for h=x^2 z, whose y partial derivative is zero

Why

For a valid homogeneous polynomial that omits one variable, SymPy represents the corresponding zero partial with a zero term at exponent (0,0,0). The coefficient-map builder attempted to place that term in the positive-degree target basis and raised KeyError, surfacing as ADAPTER_EXECUTION_FAILED.

Skipping a zero coefficient preserves the exact linear map and lets the producer return the expected degree-0 kernel. The independent verifier accepts the resulting candidate.

Validation

• python -m pytest tests/composition/runtime/test_frontier_capabilities.py -k 'graded_jacobian_syzygy or syzygy_checker_rejects' -q — 6 passed
• python -m ruff check src/jacobian/domains/polynomial/jacobian_syzygy.py tests/composition/runtime/test_frontier_capabilities.py
• git diff --check

[yuelgrace1810-ops]

Independent held-out reproduction on current main (614fdc40511d41bf88607cb0c6cc7c1e6cb91667):

• Source artifact: the exact homogenized curve coordinate H(x,y,z)=-x^3+x^2*z from Gao's 2026 tangent-sweep account of the Jacobian-conjecture counterexample.
• Request: polynomial.jacobian_syzygy.minimum_degree.compute, max_degree=0, coefficient_map_detail=CERTIFICATES.
• Two identical executions failed with ADAPTER_EXECUTION_FAILED.
• Server traceback reaches _coefficient_matrix and ends with KeyError: (0, 0, 0) while processing H_y=0.
• Independent oracle: rows/cols/rank/nullity (6,3,2,1), witness (0,1,0).

I replayed this exact polynomial against the patch in #924: producer returned the expected degree-0 kernel and the independent verifier returned VERIFIED.

Additional local validation on the PR-equivalent fix:

• 869 unit passed
• 185 domain passed
• 759 component passed, 3 platform skips
• focused old + new syzygy producer/verifier regressions passed
• Ruff lint/format and mypy passed

This confirms #924 covers another non-monomial polynomial with a zero partial, not only the existing x^2 z regression.