Improve reliability for waiting job to start before running steps.

[TingluoHuang]

• When update timeline record failed, instead of waiting for another update for the same record, we will auto retry the update with a short back-off.
• Collect how long does it normally take for the first timeline record get updated.
• Allow configure how long should we wait for the first job record updates to finish.

The change will be controlled by 3 feature flags:

• DistributedTask.EnableJobRecordUpdatedTelemetry
• DistributedTask.EnableRecordUpdateAutoRetry
• DistributedTask.FirstJobRecordUpdateWaitTimeInSeconds