[DocDB] [xCluster] ALTER DDL Abort

[nspiegelberg]

Jira Link: DB-3214

Description

When a DDL is aborted late in the transaction apply phase, we can run into a situation where the META_OP for the ALTER has been persisted in the WAL logs. The Abort happens and, with #13358, a second META_OP is written to undo the operation. This is sufficient for the single cluster use case. However, XCluster will consume this log entry and can possibly halt waiting for user-initiated synchronization (#11017).

A couple high level design options:

1. Persist the Transaction ID with the META_OP. We will soon replication the Transaction Status table, so the Consumer Master can consult that to understand if this is related to a committed transaction or one that is rolled back.
2. Inspect the contents of packets after the META_OP and only halt if an unsupported Column is used. This wouldn't happen during an Abort situation.

#2 seems like it might have a lot of performance complications and make the halting logic more complex. #1 seems more aligned with the existing design patterns for XCluster.