Summary:
Add try_wait_for and try_wait_until.

Algorithm -

We could use a doubly-linked list, but this would double the number of contended
CAS for push/post.  Instead, assume timeouts are infrequent, and just walk the list
from the head to remove nodes.

A lock bit is added to the head, which is taken only on timeout.  We assume
timeouts are infrequent.

If a concurrent post() removes the node before the timeout can remove it,
we must wait for the corresponding post() (but it is likely to arrive soon),
instead of timing out.

Reviewed By: magedm

Differential Revision: D7167894

fbshipit-source-id: ea5242098d9ccd286a72fade6292223e95c44a81