Should we just discard multiprocessing?

[goodboy]

Now that #128 is landing we don't really have a dire need for the stdlib's multiprocessing especially since it has proven much less reliable then pure trio alternatives (which I guess just means async wrapping around subprocessing?).

Reasons to keep it for now:

• windows peeps might find they need it (though CI seems to say the native trio spawner is doing just fine)
• benchmarking against the forkserver will probably be handy when we get to rolling our own for linux
• it's probably worth doing another comb through of the code to be sure we aren't missing anything critical for our own spawning stuff

Reasons to dump:

• it's super annoying to test reliably in CI
• it pickles state and sends to the child (something Drop cloudpickle dependency #133 will deprecate)
• it's currently being hacked to heck in _forkserver_override.py and if we're unlucky will have to be again with Python 3.9..
• you can't use the new debugging stuff from Wen? Multiprocessing-native debugger now! #129 with it (this is the biggest one IMO)

Relates to #84 and of course always open to opinions.