…or each event

This can't land as-is, just wanted to make sure this idea is fine, and do a try
run because I need to leave now.

This approach uses a secondary thread to manage the timeout + select.

This is not ideal, but it can be vastly improved if something is done about
rust-lang/rfcs#962.

The other apparently cleaner option, having the returned value not being an
ipc-sender, but a struct containing the sender and the Thread handle (which
would allow unparking), wouldn't work across IPC.

…o tasks

This reverts commit a9879a8.

…on the

implementation of std::sync::mpsc::Sender.

Optimal, but not ideal. I have a PR in progress to implement recv_timeout()