OperationLog replication / OpLogReceptionist should use ProtoBuf

[ktoso]

It's a pretty core element of the cluster, this one definitely should spend the effort to make the sent messages small.

Note:

Streaming in those 13kb batches (ufff, those are big but we'll solve this by a) protobuf b) compressing -- no need to keep sending the

part with the entire node all the time, since we know all those refs are of the same address, this will save A TON of space). I will make this a separate ticket though.