Implement batching of requests

[karlovnv]

Now request writer sends each request to tarantool in two tcp packets. As a result of it the number of rps is limited to maximum number of packets can be transferred by the network.

Say, if client sends 6k calls per second, the number of packets:

In: 12k p/s (to tarantool)
Out: 6 p/s (from tarantool)

Possible solutions:

• Allow user to enable Nagle algorithm
• Implement batching in connector code