Parent issue has been split up into 2.

For context:

Since transaction error messages are synced independently from the BlockExecutionData, and over gRPC which doesn't allow clear attribution for bad behavior, ingestion should not block forever if the data is not successfully downloaded. Instead, no data is written and a warning log message is logged. The API should handle this case by returning a default error message "failed". The operator can use the existing backfill admin command to add data for any blocks that were missed.

We can reuse the TxErrorMessagesCore module to handle the downloading, with retry logic similar to what we have in the engine.

These steps should be called from the processing pipeline in the downloading step.