Improve the scheduler's performance in big cluster

[nolouch]

Enhancement Task

Arm： the qps of the balance-region-scheduler should not less than 100 for 200 nodes (includes 30 tiflash).
In 6.2 version , the qps is 2op/s.

Scheduler

Placement Rule Influence

• Reduce the rule fit overhead #5522 improve 10X ==>20 ops
• schedule: use sync pool to decrease fit peer new #5553 ==> 1.2X
• region fit rule should run once. #5537
• the metrics of target filter shoule batch to report . #5538 ==> 50ops
• decrease the malloc for replace #5645

Balance region scheduler

• some filter should run once #5539 improve 1.5X ==> 30 ops
• pkg: remove reflect from slice #5555 ==> 1.5X

Batch

• One schedule() can produce multi operators

Concurrency

• handy add multi same scheduler
• Supports concurrency for each scheduler

Rule Fit

Grouping by resident label

• Label split to Location Labels and Attribution Labels. Location Labels has the keyword [Region, Zone, Host].

[bufferflies]

link #5477

[bufferflies]

ref: pingcap/log#28

[lhy1024]

It seems to be related with #3744 and #3594

[nolouch]

close due to finished. the remaining optimize may next issue.