Fix record child workflow complete mutable state stale check

[yycptt]

What changed?

• Only reload mutable state once if we detect possible stale mutable state when recording child workflow execution completed.
• This PR is a temp fix for avoiding the task infinite retry issue for 1.16 release. We should use child init event ID + version for verifying if the child actually exists or not.

Why?

• In xdc there's no guarantee that parent will have the child information and its next eventID will larger than the initiatedID in the RecordChildCompletedRequest after forced failover.
• Existing logic returns ErrStaleState, which will cause the handling logic keep loading mutable state and return and "maximum attempt exceeded" error for the record child completed API. Then task processing logic for CloseWorkflowExecution task will retry infinitely.

How did you test it?

• Existing tests. Eyeballing.

Potential risks

Is hotfix candidate?