investigate ways to combine sub-posteriors into one posterior on shear

[ismael-mendoza]

This might be necessary if we want to process > O(10^7) galaxies as we start running out of memory and the likelihood calculation becomes slow.

One way is with the gaussian posterior approximation, but unclear if for the mulpliticative bias requirement this is OK to do (see Gary's 2013 paper)

• https://arxiv.org/pdf/1311.4780
• https://arxiv.org/pdf/1304.1843