Check that cartridge-java does not create connections above limit

[vrogach2020]

We observe a strange behavior of a cluster.
It looks like java client creates lots of connections (10 times more than expected).
Cluster config:
~ 48 routers, 200 storages
~ 20 clients, 192 connections per client

Expected number of IPROTO connections in cluster ~ 5000.
Probably after some network errors number of connections increased up to 55k and remain stable for about 24 hours.

What we can do here:

• add connection count metrics to java client
• cover connection limit by unit / integration tests. Consider retrying and non-retrying configurations

[vrogach2020]

Also we should check how Netty closes sockets and if it is handled in proper way.
See https://stackoverflow.com/questions/28812858/netty-server-does-not-close-release-socket

[wey1and]

During discussions and debugging, the following problems were identified:

1. We probably have a race condition when accessing the ConnectionRegistry - we need to add thread synchronization in the AbstractTarantoolConnectionManager class.
2. It is necessary to add meaningful metrics inside the connection mechanism.
3. It is necessary to add a connection strategy, where the user can specify the following parameters: when and under what circumstances it is necessary to reconnect.
4. Add more logging to AbstractTarantoolConnectionManager.
5. We need to think about configuring the number of nioEventLoops threads. Why is their number growing and do we need to stop it?