fix: clean up inproc handler sync

[supervacuus]

I just saw that, due to the wait for the inproc handler thread in the recent changes, our inproc init benchmark regressed unnecessarily because of the poll loop's rather coarse 10ms granularity.

I took this as an inspiration to clean up the handler sync a bit:

• This PR doesn't spin on an atomic with a sleep interval each loop iteration but uses OS-level polling (pollfd on a pipe and WaitForSingleObject on an Event respectively) with a timeout. That should dramatically reduce the init wait without any wasteful sleeping on our side.
• It also extracts a significant part of the signal-handler-to-handler-thread sync into functions, so that the rather noisy sync primitives don't flood the actual behavioral code.

Would be great to pack this in before the breaking 0.13.0.

(since this is a follow-up to the unreleased #1446)
#skip-changelog

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[jpnurmi]

nice cleanup 👍