[DocDB] Use lightweight protobuf for UpdateConsensus

[spolitov]

Jira Link: DB-3243

Description

UpdateConsensus is pretty heavy and frequent RPC in our system.
It allocates a lot of memory and could release it in another thread, that does not work well with tcmalloc.
We could use lightweight protobufs to lower memory allocation and deallocation by UpdateConsensus.

[rthallamko3]

Here are some performance numbers with the fix:
Workload: java -jar nb.jar run driver=ycql yaml=v**** hosts=$NODES tags=phase:rampup cycles=4000000000..4020000000 rows_per_partition=20000 threads=1000 errors=count

Results:

Before:

v****
count = 20000000
mean rate = 62567.00 calls/second
1-minute rate = 64513.69 calls/second
5-minute rate = 63913.56 calls/second
15-minute rate = 65172.07 calls/second

After:

v****
count = 20000000
mean rate = 111018.64 calls/second
1-minute rate = 109278.04 calls/second
5-minute rate = 109752.70 calls/second
15-minute rate = 109524.26 calls/second

This yielded 80+% improvement for this workload.