feat(split): add query_child_state rpc

[hycdong]

When partition split starts, meta server will transfer split_status and partition_count through on_config_sync to replica server.
When meta partition_count = replica paritition_count * 2, primary parent replica will do different things according to meta_split_status:

• meta_split_status = SPLITTING: start split
• meta_split_status = PAUSING or meta_split_status = CANCELING: pause or cancel split
• meta_split_status = PAUSING: split has been paused
• meta_split_status = NOT_SPLIT: this pr solved

This pull request is solve lastest case (meta partition_count = replica paritition_count * 2 && meta_split_status = NOT_SPLIT), there will be two cases:

1. when primary replica register child succeed, but replica server crashed, meta server will consider this parent partition not splitting, in this case, child has been registered on meta server
2. when this parent partition is canceled, but other partitions is still canceling, in this case, child partition ballot is invalid_ballot

As above cases show, primary parent partition can not distinguish different cases, it should send a query request to meta server called query_child_state

• If child partition is ready, meta server return ERR_OK to primary, primary will make child active and update parent group partition_count
• If is not splitting, meta server return ERR_INVALID_STATE to primary, primary will ignore this case, it split has already canceled, just waiting for other paritions.

[levy5307]

What about move _replica->_primary_states.cleanup_split_states(); into parent_cleanup_split_context();

[hycdong]

It is a good recommend~ Actually, I will refactor it after I merge all function pull requests.