Endpoint::lazy_connect always tries to reconnect

[boxdot]

When a connection could not be established on first connection
attempt, an error was reported and the connection never tried to reconnect.
This behaviour is kept for Endpoint::connect and changed for
Endpoint::lazy_connect.

Motivation

The behaviour to fail on the first attempt was introduced in #413.
This makes tonic clients behave two-fold:

1. If a client fails at the first connection attempt, the connection is broken
   and it will never try to reconnect.
2. If a client connected successfully at the first attempt, but later the connection
   breaks, the client will resiliently try to reconnect.

As explained in #413 and #403, the behaviour in 1 is sometimes needed, however
there is another use case where the server we are connecting to is not yet up.
When connecting lazily, we expect the connection to
be always resilient, since we move the actual connection attempt to a later
point in time.

Solution

• Keep the error reporting behaviour in Endpoint::connect.
• Keep the resilient behaviour in Endpoint::connect_lazy.

Note

I would appreciate a proposal if there is a better way to achieve resilient initial reconnect. Also, I am not very happy to propagate the always_reconnect flag through several layers, but do not see a better way to do it.

[boxdot]

After opening the PR saw #458. Sorry for not checking. Feel free to close in case you prefer the other solution.

[LucioFranco]

Thanks for opening this, likely to go with #458