Clean up the multi-sample comparisons chapter

[LTLA]

• [x] Fix the default recommendation for the number of cells per population.
• [x] Create diagnostic MA plots for the normalization step.
• [x] Streamline extraction of metadata from the friendlier aggregateAcrossCells output.
• [x] Mention the importance of diagnostic plots for serious work.
• [x] Try switching the convenient loop to use muscat again?

[LTLA]

• [x] Add a pseudo-bulk analysis to the Segerstolpe pancreas dataset.