Root cause of error lost when retried.

[bloodearnest]

The retry logic will attempt to retry a failed db operation.

However, it does not currently preserve the original exception that caused the operation to be retried. The code that walks the exception tree to detect certain errors will then always default to the fail-through exit code of 5, rather than the specific error codes for certain known errors. This ends up having misleading messaging to users about why their jobs failed, and more tech support.

We should probably try an include the original exception in the exception tree, by passing it to the retry logic, and raise the final error from original_exception. This is both technically correct, and the existing code should work with it.