When mongos exhausts its retries on retryable errors (e.g. StaleShardVersion, sometimes SnapshotTooOld, etc.) or receives a bad status when aborting on a cleared participant between retries, it will implicitly abort the transaction and throw NoSuchTransaction to return a response with a TransientTransactionError error label to the user. Instead, the original error should always be thrown and the response should continue to generate a TransientTransactionError label.