For stale read request, we can retry DataIsNotReady replica by replica-read

[crazycs520]

Issue

1. Minimal reproduce step

In the following case, suppose the region leader is in store1.

	ca = replicaSelectorAccessPathCase{
		reqType:   tikvrpc.CmdGet,
		readType:  kv.ReplicaReadMixed,
		staleRead: true,
		label:     &metapb.StoreLabel{Key: "id", Value: "2"},
		accessErr: []RegionErrorType{DataIsNotReadyErr, ServerIsBusyErr, ServerIsBusyErr, ServerIsBusyErr},

2. What did you expect to see?

			accessPath: []string{
				"{addr: store2, replica-read: false, stale-read: true}",
				"{addr: store1, replica-read: false, stale-read: false}", // try leader with leader read.
				"{addr: store2, replica-read: true, stale-read: false}",
				"{addr: store3, replica-read: true, stale-read: false}",
			},

3. What did you see instead?

			accessPath: []string{
				"{addr: store2, replica-read: false, stale-read: true}",
				"{addr: store1, replica-read: false, stale-read: false}", // try leader with leader read.
				"{addr: store3, replica-read: true, stale-read: false}",

Currently, the replica selector logic is:

1. try store2 by stale-read, since store2 has match label, but got DataIsNotReadyErr.
2. retry leader in store1, but got ServerIsBusyErr.
3. retry store3, but got ServerIsBusyErr too.
4. failed, since no replica available.

The problem is, store2 is available, it's meet DataIsNotReadyErr, and we can use replica-read to retry store2 again.

[crazycs520]

should fix by crazycs520@2d31d6f

[crazycs520]

fix by #1181