Follow-up to #735 (perf/commitment-row-pair) addressing findings from an adversarial review. Targets the #735 branch so the fixes land with it. The intentional proof-format break (proof_sym removal / leaf-count change) is left as-is — that's by design.

Soundness of the row-pair change itself was verified clean (CPU commit ↔ GPU kernels ↔ verifier are byte-consistent; Fiat-Shamir transcript order preserved). These changes harden coverage and remove leftovers. New tests live in their own files under crypto/stark/src/tests/, matching the crate's test layout.

Test coverage (MEDIUM findings)

• tests/commitment_tests.rs (new) — direct unit tests pinning the row-pair leaf layout (R=1 and R=2) against an independent reference, wrapper agreement, commit-root consistency, and empty-input short-circuit. The layout was previously only covered transitively via full prove→verify; the GPU parity tests compare against an inline reimpl, not this module.
• tests/row_pair_opening_tests.rs (new) — two negative tests for the row-pair verify_opening_pair: a tampered symmetric trace evaluation and a corrupted Merkle authentication path must both be rejected. Removing proof_sym deleted the old "symmetric opening mismatch" rejection class; an impl that ignored evaluations_sym or the auth path would otherwise pass every existing test. (Reuses the now pub(crate) make_valid_simple_proof helper.)
• cuda_path_integration.rs — restore assert!(gpu_comp_poly_tree_calls() > 0). try_build_comp_poly_tree_gpu is dispatched unconditionally (round 2, after the parts-count branch), so it fires for the common number_of_parts == 2 (degree-3) case — it was not obsolete. The genuinely-dead parts-LDE assertion stays dropped. (Note: this test is cuda-gated, so it's only exercised on a GPU runner.)

Cleanups (LOW)

• Delete dead fn commit_plain (zero callers; main/aux paths inline commit_rows_bit_reversed + spill_tree, which are row-major and incompatible with its column-major signature — the advertised dedup never landed).
• Delete orphaned pub fn columns2rows (all callers removed by refactor(stark): unify & clean up the commitment layer #735).
• Add ProvingError::Fft and map FFTError to it instead of WrongParameter (internal FFT failure isn't a caller-supplied-parameter error; message preserved).
• Fix stale profiler label commit_composition_poly → commit_bit_reversed.
• Drop the rot-prone // = 2 comment on the local ROWS_PER_LEAF alias.

Deliberately not changed

• The debug_assert! precondition checks in commitment.rs stay debug_assert!. This is prover-side: a violated invariant yields a proof the verifier rejects, not a soundness break, so it's not worth a release-mode panic. The debug build still catches it in dev/test.

Verification

• cargo test -p stark — passes (4 commitment tests + 2 row-pair opening tests + existing suite).
• make lint — clean across all three combos (default / --no-default-features --features debug-checks / disk-spill).
• CUDA paths reviewed statically (no nvcc on the dev host); the cuda_path_integration assertion is restored by inspection.

Not addressed (out of scope per request)

• Proof wire-format compatibility / version field — the format change is the intended point of refactor(stark): unify & clean up the commitment layer #735.
• Authoritative spec (spec/main / md_spec) update for the new leaf layout — flagged for the spec owner.