change CKE to proceed rebooting immediately after draining of node is completed

[YZ775]

In the current CKE, even if max_concurrent_reboots is larger than 1, draining is conducted by serial and rebooting is conducted after draining of all drainable node is finished.
This behavior cause long down time of first drained node.

This PR changes the CKE to move rebooting immediately after draining of each node is finished.
In addition, this PR fixes behavior of canceled reboot queue entry.

[masa213f]

Thank you for fixing phaseReboot!

I checked this variable carefully, and I was convinced that this function need not return phaseReboot.
This function should only determine ops from the given information.

The caller(DecideOps()) should determine the phase. And which can be determined only by the returned ops.

Could you change the return value of this function and change the caller as following?
It's okay to change it in another PR.

https://github.com/cybozu-go/cke/blob/v1.28.0/server/strategy.go#L106-L113

	// 11. Reboot nodes if reboot request has been arrived to the reboot queue, and the number of unreachable nodes is less than a threshold.
	if ops := rebootOps(c, constraints, rebootArgs, nf); len(ops) > 0 {
		if !nf.EtcdIsGood() {
			log.Warn("cannot reboot nodes because etcd cluster is not responding and in-sync", nil)
			return nil, cke.PhaseRebootNodes
		}
		return ops, cke.PhaseRebootNodes
	}

[masa213f]

LGTM