thrift_client_pool: close bad channel before creating a new one

[jaricftw]

We see rocksplicator based applications leaking CLOSE_WAIT sockets, which is typically due to applications not closing its own sockets properly after the connection has been closed by the remote server.

To fix that, when creating a new client for a destination address, we ensure the existing thrift channel (socket) is closed, before creating a new one.

Need to test this out