Don't rely on sleeping in 0-RTT test

[Ralith]

This is unnecessarily racy, at least in principle.

[djc]

Nice! @nmittler FYI.

[Ralith]

Tangentially: a common cause of tests getting stuck until a timeout and then passing (which came up in previous related discussion) is relying on the implicit close-on-drop behavior and failing to drop something before calling wait_idle. It can be avoided robustly by closing connections explicitly.

[nmittler]

@Ralith we should probably also add a test for 0-RTT being rejected here. That's where I'm running into a timing issue in the quinn-boring PR. There is a lower-level test for rejection under quinn-proto, but it doesn't suffer from the same timing problem.