docs(firestore-bigquery-export): document failure handling and recovery

[cabljac]

Last layer of the Cloud Tasks write buffer stack. README only.

Adds a "Failure handling" section describing the sync queue, how to recover rows parked in BACKUP_COLLECTION back into BigQuery, and the known limits that have not been verified live. Adds the two new params to the settings table, syncBigQuery to the list of functions to re-export, roles/cloudtasks.enqueuer to the roles list, and updates the differences section so failed writes are described as the same buffer the extension has rather than a trigger retry.

Part of #3031.

[gemini-code-assist]

Code Review

This pull request updates the documentation in README.md to reflect the introduction of the syncBigQuery Cloud Tasks queue for buffering and retrying failed BigQuery writes. It details the new configuration parameters, IAM roles, failure handling behavior, and API surface changes. The feedback suggests a minor wording improvement in the failure handling section to clarify that a row is dropped along with the task if no backup collection is configured.

[IzaakGough]

Docs only, and I checked the claims against the code on the base branch plus the published @firebaseextensions/firestore-bigquery-change-tracker@2.2.1, firebase-admin@14.2.0 and firebase-tools. The role, both new params, the queue shape (5 attempts, 60s backoff, dispatch throttling), the trigger having no retry policy, enqueue dedupe by task id, onSuccess coming only from the queue handler in both kit and extension, the queue handler not provisioning, and the older-CLI enqueue failure all check out.

Five comments inline. All of it is from reading source, nothing deployed, and I have said where a claim rests on documented behaviour rather than something I exercised.