Change Sender to return error consistently

[vikas15bhardwaj]

Some tests fail randomly with error test is not expecting. We think its timing related as Sender always tries to reconnect every few seconds, and retries changes the error returned by Sender.

To avoid this, best for Sender is send consistent error when socket is not connected and Log the real error along with returning it to client as extra info.

[starbelly]

There's a few things I can think of here :

1. We should be logging the error when we fail to establish a conn always
2. We can have two different variations of send* (or in my current draft PR just send/3 and send/4) namely, send/3 or send/4 and send!/3 and send!/4 . This allows the user to easily pick the behavior (i.e., give me back an error or raise it)

Further, because we want to have the behavior of asynchronously connecting, which creates a scenario where by the user of the client might try to send before the conn is established, then we should store the last connection attempt error in state and clear it when we have established a conn.

I think where we might end up as far as returning a consistent error shape is with an exception struct, which is line with the above suggestions.

So, an example scenario where by the client fails to connect per a tls error but the client tries to send a message :

iex> {:ok, pid} = MLLP.Client.start_link(some_host, some_port, ...)
14:49:03.216 [error] Failed to connect ...
iex>  MLLP.Client.send(pid, message) 
{:error, %MLLP.Client.Error{status: :not_connected, error: ...}}
iex> MLLP.Client.send!(pid, message) 
** (MLLP.Client.Error) ... 

Thoughts?