safekeeper: return AppendResponse on txn commit

[erikgrinaker]

Currently, when ingesting pipelined AppendRequest, we only flush the WAL and return an AppendResponse every second. This is despite the WAL implicitly getting flushed on segment bounds (see #9688).

However, consider a concurrent writer who inserts a single row. This writer must wait up to 1 second before an AppendResponse with a new flush_lsn allows the transaction to commit.

We should consider eagerly flushing and returning an AppendResponse when the WAL contains a transaction commit. This provides the same behavior as vanilla Postgres, where a txn commit forces an immediate WAL flush.