Get queue branch ready for landing

2020-14-test5 is still being measured. It is using the queue branch which likely means the queue branch takes longer to finish measuring than master.
I can see two reasons for this:
1. Measuring on a single core machine is sometimes slower than expected. This caused the same jobs to be requeued and failed over and over. Raising the timeout for measuring a snapshot seemed to help here.
2. The measurer needs to be more robust to trial failures. Some trials died early but the measurer keeps trying to get the same snapshots that dont exist measured again and again. I think this can be fixed by having the worker signal to the manager when a trial has nothing more to measure.