Driver: Automate (re)connection logic (#81)

This PR adds several enhancements to Driver connection logic:
* Driver (re)connection attempts now have a default timeout of around 10s.
* The driver will now attempt to retry full connection attempts using a user-provided strategy: currently, this defaults to 5 attempts under an exponential backoff strategy.
* The driver will now fire `DriverDisconnect` events at the end of any session -- this unifies (re)connection failure events with session expiry as seen in #76, which should provide users with enough detail to know *which* voice channel to reconnect to. Users still need to be careful to read the session/channel IDs to ensure that they aren't overwriting another join.

This has been tested using `cargo make ready`, and by setting low timeouts to force failures in the voice receive example (with some additional error handlers).

Closes #68.

src/handler.rs

@@ -203,7 +203,7 @@ impl Call {
         let (gw_tx, gw_rx) = flume::unbounded();
 
         let do_conn = self
-            .should_actually_join(|_| Ok(()), &tx, channel_id)
+            .should_actually_join(|_| (), &gw_tx, channel_id)
             .await?;
 
         if do_conn {
@@ -218,6 +218,14 @@ impl Call {
                 .await
                 .map(|_| Join::new(rx.into_recv_async(), gw_rx.into_recv_async(), timeout))
         } else {
+            // Skipping the gateway connection implies that the current connection is complete
+            // AND the channel is a match.
+            //
+            // Send a polite request to the driver, which should only *actually* reconnect
+            // if it had a problem earlier.
+            let info = self.current_connection().unwrap().clone();
+            self.driver.raw_connect(info, tx.clone());
+
             Ok(Join::new(
                 rx.into_recv_async(),
                 gw_rx.into_recv_async(),