Consider BatchRegionHeartbeat

[overvenus]

Currently, we use region heartbeat to do scheduling, one region one heartbeat stream. For every heartbeat stream, gRPC starts a goroutine, it works fine on small clusters, however, it may cause too much memory and CPU on large clusters. A straightforward solution is introducing a new rpc, BatchRegionHeart, it batches multiple region heartbeats into one message, I think it should save memory and CPU, and is a better way to do batch handling region heartbeats, #1448 .

[bradyjoestar]

And currently leader broadcast regions info to follower，
could every follower get batch region's info from tikv store directly and cached it and its hash
and then wait a moment try to ask leader that have it seen this hash，this may release the pressure of leader too？:grin: