Consolidate duplicate distance searches and parallelize the search engine

[ciaranra]

Stacked on #415 (base code-distance-rust); retarget to dev once that merges.

Consolidation

PECOS had two independent implementations of the same weight-increasing search: calculate_distance/find_shortest_logicals in pecos-qec/src/distance.rs, and has_undetectable_logical/compute_distance in fault_tolerance/stabilizer_flip_checker.rs with its own enumeration helpers.

Their predicates were verified equivalent -- both test "commutes with every stabilizer generator AND anticommutes with at least one configured logical Z or X" -- so the checker methods now delegate to the shared engine. A new has_logical_error_at_weight covers the exact-weight case the checker needs. The checker's existing unit tests are unchanged and act as the regression guard.

The combinations/pauli_product/build_pauli_string helpers are deliberately retained: analyze_weight still needs them for configurable X/Y/Z subsets that WeightedPauliIterator does not express.

Parallel search

The per-weight candidate scan now runs on rayon, partitioned across support combinations with Pauli assignments enumerated serially within each support. Output is bit-identical to the serial path rather than merely equivalent: reduction is on enumeration index, so calculate_distance returns the same operator the serial search returned and find_shortest_logicals returns the same vector in the same order. Tests cover both the serial and the parallel branch.

Parallelism engages per weight, not per code, so a large code's cheap low weights stay serial.

Measured results

PARALLEL_CANDIDATE_THRESHOLD = 65_536 candidates at a single weight, derived from a sweep rather than chosen by intuition. Below roughly 22k candidates parallelism loses -- forcing the toric [[18, 2, 3]] weight-3 tier (22,032 candidates) parallel made that search 4.6x slower. Above roughly 193k it stops engaging where the time is spent -- the color [[17, 1, 5]] search is dominated by its weight-4 tier (192,780 candidates), and a threshold past that erased the speedup entirely. The chosen value sits near the geometric centre of that window.

Benchmark	Serial	Parallel	Effect
calculate_distance five-qubit [[5,1,3]]	7.70 us	8.46 us	9.9% slower
calculate_distance Steane [[7,1,3]]	15.04 us	16.83 us	11.9% slower
calculate_distance color [[17,1,5]]	41.7 ms	8.6 ms	4.8x faster
find_shortest_logicals delta=1, color [[17,1,5]]	2.77 s	215 ms	12.9x faster

This is a trade, not a free win: microsecond-scale searches pay about 10% for the per-weight threshold check, while the searches long enough for a user to wait on improve by 5-13x. Small-code numbers were confirmed with an A/B/A run after an initial measurement proved to be machine drift.

Adds crates/benchmarks/benches/modules/code_distance.rs; no distance benchmark existed before this.

Verification

• cargo test -p pecos-qec -p pecos-quantum: 22 test groups, zero failures
• just build-debug, just lint (with new files staged so pre-commit sees them): clean
• uv run --frozen pytest on the stabilizer-code binding suites: 28 passed