Other than occasional flukes here and there the biggest offender seems to be
*after-suspend* family of subtest.

The most representative scenario looks like that:
1. schedule something on chamelium
2. got to sleep
3. scheduled thing triggers
4. wake up
5. check whether we see the changes reflected through DRM and pass
6. exit triggring chamelium cleanup

During 6 we have a failed RPC because network is not up yet after waking up,
which overwrites the test results to WARN.

Solution: introduce chamelium_wait_online(int timeout) that pings the device
and bails out after timeout. Always call it after waking up.

Everything else looks like sporadic network issues, but we have seen it just a
few times in the last months. We can investigate them further once we will get
rid of the main source of the noise.

Impact on users: none, it's a CI/test issue.
Impact on testing: potentially huge, as some of the tests may leave us with
chamelium ports unplugged adding to the flip-flopping of the 2x tests.

Bumping priority to high, as it is easy to solve and important for keeping the
CI noise down.