pythonGH-102796)

This test can fail unnecessarily. In the test we wait for events on two
file descriptors. This is done in a single call to select.epoll's poll()
function. However, it is valid for the OS to return only one event via
poll() and the next via a subsequent call to poll(). This rarely
happens, but it can cause the test to fail despite properly functioning
polling.

Instead, we poll a second time when necessary.
(cherry picked from commit c9ecd3e)

Co-authored-by: Kevin Krakauer <kevinGC@users.noreply.github.com>