Make a higher level "chi2 table" like API

[wilsonmr]

So many of the tables in results and now closure are small variations on a theme. I was wondering if we could shorten the code a bit by making some kind of table API which allows tables to be declared in runcard somehow

tablespec:
  index: {'experiments': 'datasets'}
  columns: {`fits`: {`ndata`, `central`}}
  show_total: True

actions_:
 - chi2_table_action

maybe we could even be more adventurous and have cross compatibility for anything which can be tabled in this way:

tablespec:
  index: {'experiments': 'datasets'}
  columns: {'fits': {'ndata', {'chi2': 'central'}}}
  show_total: True

actions_:
 - generic_table_action

Not sure if that would even be useful however.. Probably the first thing would be sufficient to start

I haven't thought that through really but I think it could work? I'm sure @Zaharid already mentioned it in #447

specifically: #447 (comment)

[Zaharid]

I think the generic thing would be a nighmare of corer cases. E.g. what does show_total mean generically?

Note also that the first one doesn't cover what we do now. The total chi2 for a given experiment is computed separately taking into account the correlations per dataset.

[wilsonmr]

Perhaps, even if not, it isn't clear if it would be useful. But I guess I was thinking that anything which goes into this table can be expressed in the same form as a chi2, and so can be contained in an object which is similar to abs_chi2_data (some can't have replica data - although they could be bootstrapped)
then calculating the totals is just collected over all data and summing (making sure that the values saved in abs_chi2_data containers are not per data point) this works for phi, bias, chi2 (all the things I was thinking about)

the table API would not calculate the total internally, it would taken from another action which does what I said above

The table index could then be any acceptable grouping of data

[wilsonmr]

anyway I'm happy to focus on the first thing with just chi2 to start and see how well it comes together before worrying about extending it to other estimators