Retry works by checking if an non-protocol error happened during the call, if so an attempt to take a write lock on the client is made and a new client attempted
if both operationare succesfull the old_client is substituted with the new one.
Timeout is a single parameter for connect,read and write TCP timeouts, special handling is required for the following situation: cannot set a timeout with a proxy and cannot set timeout with multipl SocketAddrs
Since configurations parameter grow, a Config struct with a builder has been introduced
Some new errors variant has been introduced
Errors in reading input in the reader thread now warns other eventual threads with ChannelMessage::Error
Add github actions equivalent to travis cause travis is slow
This new model is based on the idea of having only one thread monitor the socket and then
distribute the responses to every other thread that's waiting.
Whenever a request is made, a new mpsc channel is created and added to an internal map.
Then, if there are not other threads waiting for pending responses, that thread will
become the "reader thread" and monitor the socket until it receives the response it was
waiting for. In the meantime, if it sees responses coming on the socket with different
ids, it will lookup the correct Sender from the internal map and send that message to
the thread that created the request in the first place.
The other threads will be waiting trying to read from their Receiver, where they will
either get the message in response to their request, or a signal to wake up and
become the new "reader thread".
