fix(client): detect connection closes as pool tries to use

Currently, if the remote closes the connection at the same time that the
pool selects it to use for a new request, the connection may actually
hang. This fix will now more allow the keep-alive read to check the
socket even when the `Conn` think it's busy.

If the connection was closed before the request write happened, returns
back an `Error::Cancel`, letting the user know they could safely retry
it.

Closes #1439

src/client/dispatch.rs

@@ -85,7 +85,7 @@ impl<T, U> Drop for Receiver<T, U> {
         // - Err: unreachable
         while let Ok(Async::Ready(Some((_val, cb)))) = self.inner.poll() {
             // maybe in future, we pass the value along with the error?
-            let _ = cb.send(Err(::Error::new_canceled()));
+            let _ = cb.send(Err(::Error::new_canceled(None)));
         }
     }
 

src/client/mod.rs

@@ -243,11 +243,7 @@ where C: Connect,
                 },
                 Err(_) => {
                     error!("pooled connection was not ready, this is a hyper bug");
-                    let err = io::Error::new(
-                        io::ErrorKind::BrokenPipe,
-                        "pool selected dead connection",
-                    );
-                    Either::B(future::err(::Error::Io(err)))
+                    Either::B(future::err(::Error::new_canceled(None)))
                 }
             }
         });