v0.4.2 — durable retry logging

Fixed

• Durable retry handler and branch advancer now log caught exceptions instead of swallowing them silently (#1). Previously, any exception thrown inside an agent's prompt() call (or any tool side-effect, broadcast dispatch, etc.) was caught and either rescheduled for retry or marked as a terminally failed branch with no entry in the application log, failed_jobs, or anywhere else — producing symptoms identical to transient LLM API failures and making the root cause invisible without reading the package source.
  • DurableRetryHandler::scheduleRunRetryIfAllowed and scheduleBranchRetryIfAllowed now emit Log::warning('Durable swarm {step,branch} failed — scheduling retry.', [...]) before scheduling, with run_id, retry_attempt, max_attempts, next_retry_at, exception class, and message (plus branch_id / agent_class on the branch path).
  • DurableBranchAdvancer now emits Log::error('Durable swarm branch failed — retries exhausted or non-retryable.', [...]) before calling markBranchFailed, with the same fields. The run-level terminal failure path already rethrows via DurableStepAdvancer, so failed-jobs / Laravel exception handler observability is already present there; the branch path was the silent one because its catch block returns normally after the failure.
• Both DurableRetryHandler and DurableBranchAdvancer now accept a constructor-injected Psr\Log\LoggerInterface. Container auto-resolution handles existing wiring; no service-provider changes required for application code.