Fix flaky test_oversubscribing_leases

[fjetter]

This should finally fix #3717 by introducing two changes

• A new method get_value to observe the state. This eliminates any timing based assumption on the oversubscription but allows us to wait until that condition is given for certain
• For some reason, the PeriodicCallback start/stop within the test did not work as expected. I got stop the callback, I could start it again but for whatever reason it was not doing anything. That left the entire test in more or less random state and I'm wondering why it worked in the first place. To eliminate the randomness I control the periodic callback via the refresh_leases variable now instead of start-/stopping it. Also, I explicitly verify that the internals do what they are supposed to do by capturing the logs. Has anybody else encountered a PC behaviour like this before? Is start/stop/start something we are not supposed to do with PCs?

I couldn't reproduce the race condition in the first point but I am now more confident that the test is doing what it was intended to do when running it locally and I don't see any more room for race conditions atm.

Sorry for the flaky test, I hope this fixes the issue now

cc @mrocklin

[mrocklin]

Thank you for following up on this @fjetter ! I appreciate it.