Delta chi2 table by group

[wilsonmr]

So I think due to the timing of when #1151 got merged, something broke and it's to do with collecting data over fits with use_fitcommondata True. For some reason it doesn't get properly resolved and all the commondata files point at those from the first fit. I think it's because we now implicitly parse the datasets from the outer namespace and for some reason this caches the first result somewhere, I guess it's because produce_data doesn't set fit in the context when parsing the individual datasets but I'm not really sure.

After a long time if messing around with this I got rid of the old functions which largely were uninteresting and unused and just fixed delta chi2 table which is what we show in the paper. Adding the ability to use an arbitrary grouping turned out to be a lot more effort than I thought (and probably than it's worth but there we go..)

There is some annoying internals here which really ought to be fixed. Now that we can used python commondata we should really use it to generate the level 1 data and then save it in a sensible format so this isn't required.

Anyway, you can now group by either process or experiment: https://vp.nnpdf.science/XbuRheXoRU60J_C68x3hyg==

@tgiani I'll hand this over to you, but I realise you've got a lot of closure related PRs now so you might want to delegate some tasks.