Only report Oban job failures to Sentry after all retries are exhausted

Hive now only reports Oban background job failures to Sentry after a job has exhausted all of its retry attempts, instead of reporting every failed attempt. This removes the Sentry alert noise caused by retryable errors (for example, transient timeouts in the GitHub issue classification worker) while jobs still have retry budget left. Jobs whose attempts are fully exhausted continue to be reported with the same Oban metadata. Operators who want per-attempt reporting back can opt in by setting SENTRY_OBAN_REPORT_RETRIES=true. Self-hosted Helm deployments get the quieter default through the Helm chart values, and no migration is required.