Keep local activity scheduleToClose deadline across replay

[sangkyoonnam]

What was changed

LocalActivityFailedException now keeps the originalScheduledTimestamp it is built with, and the timer-backed retry in SyncWorkflowContext passes that marker value to the next attempt instead of the System.currentTimeMillis() captured when workflow code ran. When the marker has no metadata (-1) or no firstSkd (0), it falls back to the captured value as before.

Why?

On replay, the captured value is the replay clock, so each replay restarted the ScheduleToCloseTimeout budget and a failing local activity could keep retrying until MaximumAttempts. sdk-core and sdk-go both keep the original schedule time across replay. Details and a reproduction in #3106.

No SdkFlag: the recorded markers and backoffs still drive replay, and the change only affects the deadline the local activity worker computes for a live attempt.

Checklist

1. Closes Local activity scheduleToClose budget restarts when a timer-backed retry runs on replay #3106

2. How was this tested:
   LocalActivityScheduleToCloseAcrossReplayTest forces a full-history replay on every workflow task (setStickyQueueScheduleToStartTimeout(Duration.ZERO)) and asserts that every local activity marker carries the first attempt's firstSkd, plus RETRY_STATE_TIMEOUT before MaximumAttempts. Without the fix each marker carries a new firstSkd and the run ends at RETRY_STATE_MAXIMUM_ATTEMPTS_REACHED. io.temporal.workflow.activityTests passes (117 tests).

3. Any docs updates needed?
   No.