Parallelize automatic section-mask candidate evaluation and support merging multiple winning candidates

[0hmX]

Motivation

• Increase throughput and robustness of the automatic section-mask search by evaluating candidates in parallel and allowing multiple non-overlapping beneficial candidates to be merged into a single section mask.

Description

• Added CandidateEvaluationResult and isCandidateEvaluationResult, refactored candidate evaluation into a chunked/parallelized evaluation pipeline that aggregates per-candidate timings and results.
• Batch-evaluates candidates in chunks of PARALLEL_CANDIDATE_SOLVE_LIMIT (default 4) and collects solved, replayed finalMaxRegionCost values for comparison.
• Chooses multiple non-overlapping candidates by improvement, merges their port masks, replays the merged solution, and updates the best mask; replaced single winningCandidateLabel/family with winningCandidateLabels and winningCandidateFamilies.
• Added new TinyHyperGraphSectionSolverOptions.PARALLEL_CANDIDATE_SOLVE_LIMIT and DEFAULT_PARALLEL_CANDIDATE_SOLVE_LIMIT, exposed the option in applyTinyHyperGraphSectionSolverOptions and getTinyHyperGraphSectionSolverOptions, and set default properties on section solvers.

Testing

• Ran project build via yarn build to ensure TypeScript compilation succeeded and the bundle compiles correctly, and it passed.
• Ran unit test suite via yarn test to validate behavior of the pipeline changes, and tests passed.

---

Codex Task