Update for latest IO library changes.

The new `IO` library has an `IO.Action.NotHandled` action,
which `IO.CoreEngine` uses to indicate that it doesn't own the event.

It is now the responsibility of `TCP.Engine` to handle
such situations instead of making `IO.CoreEngine` handle it.

src/TCP.Engine.savi

@@ -85,14 +85,28 @@
 
       | IO.Action.Read |
         if Platform.is_windows (
-          None // TODO: @_windows_complete_reads(arg)
+          @_finish_read_windows(event.arg) -> (yield action)
         |
           @_pending_reads_unix -> (yield action)
         )
 
       // TODO: windows complete writes, flush-after-mute (pending writes logic from Pony)
       // | IO.Action.Write |
       //   ...
+
+      | IO.Action.NotHandled |
+        // If the underlying engine didn't want to handle this event, it means
+        // this is likely a later connection attempt after we already had one.
+        // If we have more connection attempts pending, we'll unsubscribe it.
+        if (@_pending_connect_count > 0) (
+          @_pending_connect_count -= 1
+          _FFI.pony_asio_event_unsubscribe(event.id)
+        |
+          // If we have no more pending connection attempts, pass the buck to
+          // the caller to figure out what needs to be done with this event.
+          // Maybe they have another engine sharing the same IO actor, perhaps.
+          yield action
+        )
       |
         yield action
       )