QED fitting performance improvements: either a photon-pdf resource or vp-setupfit pre-computation of fiatlux

[scarlehoff]

It would be ideal if we could run fiatlux during vp-setupfit so that then we can run the fits in parallel.

then the possibilities are two:

1. Either we store the values in the server so someone running the same PDF-theory combination (which in practice will happen many times) can just reuse the existing values.
2. Or, if it is fast enough, we can try parallelizing the computation of the photon over replicas with vp-setupfit (vp-setupfit --pre-compute-photon or something)

[scarlehoff]

Added to this issue, as reported by @RoyStegeman
fiatlux fails to converge for some replicas, we'd want to catch this earlier (perhaps a timeout?)

Making fiatlux itself faster would be helpful here since then one can put a very aggressive timeout.