Use golang's context to cancel query, rather than proactively check a killed/canceled flag

[lance6716]

Enhancement

as title.

pros:

• golang runtime and libraries have adopted the style that context.Context is the first parameters in function arguments. We don't need to proactively check the query is killed/canceled frequently. In many functions from official SDK or 3-rd party libraries, invalid context will cause the function return error and break the workflow. In fact, we often forget to proactively check the killed/canceled flag and have caused many problems like can't kill query for CHECK_CONSTRAINTS #58080 non-dist-task fast-reorg ADD INDEX can't be canceled timely #56017
• after golang 1.20, we can return specific error to https://pkg.go.dev/context#CancelCauseFunc , so the reasons of

  tidb/pkg/util/sqlkiller/sqlkiller.go

  Lines 71 to 85 in 0ffac36

  	func (killer *SQLKiller) getKillError(status killSignal) error {
  	switch status {
  	case QueryInterrupted:
  	return exeerrors.ErrQueryInterrupted.GenWithStackByArgs()
  	case MaxExecTimeExceeded:
  	return exeerrors.ErrMaxExecTimeExceeded.GenWithStackByArgs()
  	case QueryMemoryExceeded:
  	return exeerrors.ErrMemoryExceedForQuery.GenWithStackByArgs(killer.ConnID.Load())
  	case ServerMemoryExceeded:
  	return exeerrors.ErrMemoryExceedForInstance.GenWithStackByArgs(killer.ConnID.Load())
  	case RunawayQueryExceeded:
  	return exeerrors.ErrResourceGroupQueryRunawayInterrupted.FastGenByArgs("runaway exceed tidb side")
  	default:
  	}
  	return nil

  can be passed in context way

cons:

• the builtin style of CancelCauseFunc requires to use https://pkg.go.dev/context#Cause to find the error, but we are more familiar with ctx.Err(). There should be some work to rewrite the usage of ctx.Err()
• the builtin https://pkg.go.dev/context#WithCancelCause uses a lock to guard the cancel cause. So if multiple goroutines access it, it's slower than the killed/canceled flag which is an atomic variable. Given these 2 drawbacks, we may want to implement our own WithCancelCauseFast function

[lance6716]

tidb/pkg/server/server.go

Lines 959 to 965 in d6b313f

	conn.mu.RLock()
	cancelFunc := conn.mu.cancelFunc
	conn.mu.RUnlock()
	if cancelFunc != nil {
	cancelFunc()
	}

context is already canceled, but there still be many reports of query stuck. Maybe they didn't use correct context