OK-5110 make stop synchronous

[GerryMandell]

Summary

Make avd module's stop() action synchronous: wait for the run-avd process to actually exit before returning, instead of fire-and-forget on SIGTERM.

Problem

avd.py's stop() previously sent SIGTERM and returned immediately:

os.kill(running_avd["pid"], signal.SIGTERM)
self.result["changed"] = True
return self.result

Since /opt/orka/bin/run-avd is a bash script that manages child processes (the emulator and socat relay), it can take a few seconds to handle SIGTERM and exit cleanly. Callers that immediately query AVD state (via gather_avd_facts or pgrep) see the AVD as still "running" — even though the stop call succeeded.

This caused validation steps in our CI workflow to flake, and would surface as a confusing "AVD is running" state in any subsequent task that depends on a stop having taken effect.

Fix

After sending SIGTERM, poll the process with kill(pid, 0) until it exits or the 30s timeout expires. If the process doesn't exit in time, fail the task explicitly so callers know the stop didn't take effect.

Test plan

• Run avd.yml -e desired_state=stopped against a running AVD — confirm task only returns after the process is actually gone (pgrep -f run-avd | grep <name> returns nothing immediately after)
• Run avd.yml -e desired_state=stopped against a stuck process (e.g. one trapping SIGTERM) — confirm the task fails with a clear timeout message rather than reporting success
• Run the full deploy → stop → validate cycle in CI — confirm validation no longer races

[spikeburton]

nit - I do not think this comment is needed. This is a common way to query if the process is running, in fact we use it in the run-avd script itself here

But it doesn't add too much noise, so I am fine to leave it, upto you

[GerryMandell]

Removed that comment.