Make the Local TSO Allocators in the same DC to start the election first

[JmPotato]

Feature Request

Describe your feature request related problem

For now, we use (*AllocatorManager).PriorityChecker() to make sure a Local TSO Allocator is elected from the corresponding DC as possible. However, when a new cluster is started or some PD instances in the same DC are joined, there is the possibility that the Local TSO Allocator is played by a PD node from another DC. That's because the priority is guaranteed by PriorityChecker to check periodically whether the current Local TSO Allocator is from its DC correctly, which has some lagging depending on the inspection period.

pd/server/tso/allocator_manager.go

Lines 784 to 791 in b157862

	// PriorityChecker is used to check the election priority of a Local TSO Allocator.
	// In the normal case, if we want to elect a Local TSO Allocator for a certain DC,
	// such as dc-1, we need to make sure the follow priority rules:
	// 1. The PD server with dc-location="dc-1" needs to be elected as the allocator
	// leader with the highest priority.
	// 2. If all PD servers with dc-location="dc-1" are down, then the other PD servers
	// of DC could be elected.
	func (am *AllocatorManager) PriorityChecker() {

Describe the feature you'd like

We need to make sure the PD nodes from the corresponding DC become its Local TSO Allocator as soon as possible rather than being corrected by the PriorityChecker later, which should make the cluster provide more stable and fast TSO service to TiDB.

Describe alternatives you've considered

Maybe we can trigger a DC's Local TSO Allocator election earlier than other PD nodes in the other DCs.

[okJiang]

/close ref #2759 (comment)