etcd can't serve serializable read request when the quorum isn't satisfied

[ahrtr]

This issue applies to both 3.5 and the 3.6 (the main branch). I do not get time to verify previous versions.

Steps to reproduce this issue:

1. Start an etcd cluster with 3 members, and add some K/V data;
2. Stop all the members;
3. Start only one member;
4. Try to issue a serializable request using command something like etcdctl --endpoints http://127.0.0.1:32379 get k1 --consistency="s", then you will get the error like below,

$ etcdctl --endpoints http://127.0.0.1:32379 get k1 --consistency="s"
{"level":"warn","ts":"2021-12-09T06:18:40.477+0800","logger":"etcd-client","caller":"v3/retry_interceptor.go:63","msg":"retrying of unary invoker failed","target":"etcd-endpoints://0xc00000a3c0/127.0.0.1:32379","attempt":0,"error":"rpc error: code = DeadlineExceeded desc = context deadline exceeded"}
Error: context deadline exceeded

The root cause of this issue is that etcd waits for the ready notification infinitely on startup, please see serve.go#L96. But when the quorum isn't satisfied, the etcd instance will never get the notification.

The etcd instance should NOT wait for the ready notification infinitely; otherwise, it will never be able to serve serializable read request when the quorum isn't satisfied.

I will submit a PR soon, and any feedback is appreciated.

@serathius @xiang90 @gyuho @ptabor