Latent Variable Support for Analysis Summed

[Jammy2211]

The following PR implements a hotfix to get latent variables working for summed Analysis objects:

#1084

We need to flesh out the PR so that latent variables fully use all summed Analysis features, for example when the model has free parameters that vary across the analysis objects.

However, it may just be that no source code changes are required and its up to a user to customize their compute_latent_variables correctly.

We can discuss this in the new year.

[rhayes777]

I suppose it's possible that each of the analyses that have been combined may have its own latent variables and so we need to compute latent variables for each separate analysis? Perhaps we need to call this for each analysis with a distinct type.

The same could arguably apply to visualisation. At the moment we're favouring the first analysis. The avoids repetition where all analyses are the same but if a user is composing multiple distinct analyses they will be missing visualisation.

I find these combined analyses hard to reason about - sometimes I wonder if we've got the abstraction right!