Abort execution if any of the prefetchers failed

[ayllon]

This should "fix" the cases where sourcextractor++ stays running doing nothing. It does not fix the underlying reason, but at least it gives an explanation and exits.

[mkuemmel]

Okay with me.

Unfortunately it does not help with the particular dataset I am working on right now. That behaves still quite erratic.

[ayllon]

Ok, I have added the same check on the multi-threaded measurement. It should catch when the thread pool dies if the prefetcher doesn't.

[mkuemmel]

The last changes did not help. But the problem is more complex:

• the dataset runs, without fitting, sometimes through with v0.16 (once out of ~4 tries), otherwise one processor stays at 100% without progress;
• v0.17 without fitting does not progress after 10% or 20% with one processor runnning at 100%;
• with model fitting it does not finish, sometimes one processor is at 100%, sometimes not;

This is really kind of bizarre. I checked the 'old' Euclid dataset that bothered us last year, but that seems to run with both v0.16 and v0.17.