Some testbed failures result in infinite retry loops

[Paride Legovini]

Testbed failures are not actual test failures, and there's logic in the autopkgtest-cloud worker script to detect them and retry the test (it is put back in the queue). However there are cases where some tests can fail in a way that is detected as a testbed failure. This results in an infinite retry loop which keeps the workers busy.

We should improve this retry logic, perhaps limiting the number of retries then a testbed failure happens.