Fix potential spin loop when a read on the Wayland socket initially fails

[lilyyy411]

The current handling of reading the Wayland socket lends itself to accidentally causing a spin loop when there are many events or there are multiple threads reading/writing events to the Wayland socket. read_without_dispatch without a corresponding call to dispatch_inner_queue is especially prone to accidentally causing a spin loop.

The thing is that we don't need a read_without_dispatch here and, in fact, we want the behavior of read because we want the events to leave the backend's inner queue and enter the EventQueue so they can be dispatched through the application.