Schedule triage runs more intelligently

[timokau]

Currently it just runs whenever requested (when a relevant state was set) or after one day. It might also run whenever reviewers become available (maybe conditioned on the cached number of needs_reviewer PRs) (#71) or whenever we predict time-out action to be necessary. As a result, reviewer would be assigned more quickly and timeouts would happen more timely.

[timokau]

I could imagine something like

await has_reviewer_available()

where has_reviewer_available somehow blocks (or yields the CPU) until the caching mechanism in #71 has found an available reviewer.