gc stale peer

[siddontang]

See https://github.com/pingcap/tikv/blob/master/src/raftstore/store/store.rs#L356

We don't handle following case now:

nodes 1, 2, 3, 1 is leader, 2 is isolated.
1 adds 4, 5, 6, Now assume 4 is leader, and 4 removes 1, 3 first, then remove 2.
When 2 re-joins the cluster, it will send vote to 1, 3, but 1, 3 will ignore this message, and 2 can't know it is already stale.

To handle this, if a peer has no leader for a long time (e.g, 10m, set in configuration), it will ask pd whether it is stale or not. We can save a last check time and check this in every region tick.
To ask pd, we can use pd GetRegion with region start key directly, if the returned region doesn't contain this peer, we can think the peer is stale.

Notice:
The GetRegion may return another next region if splitting. we should handle this case.