The etcd client does not use auto-sync and fails when the PD cluster address changed

[sdojjy]

reproduce step

Start a TiDB cluster with 3 PDs ① ② ③ and a ticdc connected
Scale-out 3 more PDs ④ ⑤ ⑥
Wait 31 seconds
Scale-in the original PDs ① ② ③

ticdc is restarted.

[asddongmen]

TiCDC may shutdown or get stuck in this scenario.
It was stuck in this function:

// campaign to be an owner.
func (c *captureImpl) campaign(ctx context.Context) error {
	failpoint.Inject("capture-campaign-compacted-error", func() {
		failpoint.Return(errors.Trace(mvcc.ErrCompacted))
	})
	// TODO: `Campaign` will get stuck when send SIGSTOP to pd leader.
	// For `Campaign`, when send SIGSTOP to pd leader, cdc maybe call `cancel`
	// (cause by `processor routine` exit). And inside `Campaign`, the routine
	// return from `waitDeletes`(https://github.com/etcd-io/etcd/blob/main/client/v3/concurrency/election.go#L93),
	// then call `Resign`(note: use `client.Ctx`) to etcd server. But the etcd server
	// (the client connects to) has entered the STOP state, which means that
	// the server cannot process the request, but will still maintain the GRPC
	// connection. So `routine` will block 'Resign'.
	return cerror.WrapError(cerror.ErrCaptureCampaignOwner, c.election.campaign(ctx, c.info.ID))
}

[asddongmen]

ref: pingcap/tidb#42643