exceptions reported to stderr when they propagate past forkIO
It's not entirely obvious what to do with exceptions that do not get handled within a
forkIO however reporting them on stderr (or on Windows popping up a message dialog) does not seem right.
We do not have other cases where errors are logged to stderr. The only such case is an exception terminating Main.main (and that's special because it terminates the whole process). If it is vital that someone do something with exceptions in forkIO threads then they should be propagated to another thread, in the worst case the main thread.