Speculative fault-space exploration

[symbiont-stevan-andjelkovic]

Currently lineage-driven fault injection (LDFI) returns a set of faults, which we inject in the next concrete run, which gives us a new trace and thus new constraints for the LDFI making it better at guessing what the next set of faults should be, and so on. Clearly, as the set of constraints grows LDFI will get slower.

One possible optimisation that could help in this situation is to have LDFI return several sets of faults instead of just one, and thus do several concrete runs of the test per run of LDFI. We should be able to do this today with minor changes.

An obvious next step would be to run these concrete runs in parallel. This would require more work though, e.g. making the scheduler be able to keep track of several runs at the same time, and it would also put more strain on the db so we probably want to switch over to using the logger component which batches writes to the db. Should be done after #70, #72, and possibly #65.