v1.107.0

Added

• catch parameter on Bench.optimize(), forwarded to optuna.Study.optimize(): a trial whose worker raises one of the given exception types is recorded as FAILED and the study continues with the remaining trials, instead of one raising trial aborting the entire study. The default () mirrors Optuna's own default and preserves the existing fail-fast behaviour exactly. ParametrizedSweep.to_optimize() already forwards **kwargs, so it picks up catch with no change. A raising worker leaves no committed sample-cache entry (both the serial and parallel paths raise before cache.set), so FAILED trials cannot poison the cache. Coverage in test/test_optimize.py::TestCatch.