Guarantee history task execution

[yycptt]

What changed?

• New Single cursor mode for transfer and timer processor.
  • When enabled, only one queue processor (instead of one for active and n for standby clusters) will be started and handling tasks for both active and standby namespaces.
  • When enabled, queue processor won't register cluster metadata callback for (re-)creating standby processors.
  • When enabled, history engine will still register namespace change callback since the callback contains some logic for namespace migration. However, lock/unlock task processing and failover namespace call inside the callback will be a no-op (so no more failover queue).
  • When enabled, standby task execution logic will use current cluster time.
• History task executor now also returns a metrics scope that should be used for emitting metrics for task processing. The reason is that active & standby tasks need to use different metric scope, but whether task is processed as active or passive is unknown before calling the executor.
• Remove unused time source in timer sequence.
• This PR depends on some code refactoring on xdc history resender so that resender is no longer bind to a certain remote cluster, which is still WIP.

Why?

• Short term implementation for guarantee task execution and ensure no task loss during namespace failover. In long term, queue processor will be re-written using new multi-cursor queue components.

How did you test it?

• Tested locally using go canary running in a global namespace with both single cursor flags enabled

Potential risks

Is hotfix candidate?