Add MCMC combine method and improve shuffle behaviour when combining results

Cherry picked results from !945 (merged)

Note: this will change the default behaviour when combining both MCMC and nested samples. Instead of shuffling the results together, they will be unshuffled enabling converge checks of the individual chains.

• Add option to shuffle posteriors when combining (default to false)
• Add mcmc sample combine

changed milestone to %1.1.2

To clarify the change, here are some plots.

Current behaviour

If you run pesummary (or just plot the posterior chain) for a single dynesty run. You will get a plot which shows the evolution of samples as the sampler climbs the likelihood surface:

But, if you plot the samples from multiple combined samples from dynesty and the current master you will get something that looks like this:

The samples get shuffled and you lose the structure.

Proposed behaviour

With this MR, you will get the same plot if you run a single sampler. But, if you merge multiple samplers runs together you'll get:

This is an improvement because: a) The current behaviour is inconsistent between single and multi-chain runs leading to user confusion b) Seeing the individual chains enables an assessment of convergence and the identification of "bad chains". c) Downstream users may still shuffle their posteriors if they want to remove the structure.

changed title from Cherry picked results from !905: to Add MCMC combine method and improve shuffle behaviour when combining results

changed the description

approved this merge request

added 3 commits

• 086ffce4...fef914d0 - 2 commits from branch master
• 27fa1e48 - Merge branch 'master' into result-improvement

Compare with previous version

approved this merge request

approved this merge request

added Sampling label

merged

mentioned in commit fea22763