Reorganize pseudobulk analysis functions

[yihming]

Modify pseudobulk function:

• Rename sample to groupby
• Rename cluster to condition
• Return a MultimodalData object, and don't write to the original data object
• Enforce non-numeric attributes set in attrs to be categorical in the pseudobulk data object

Modify deseq2 function:

• It takes 2 backend subfunctions run_pydeseq2 and run_rdeseq2. Default is run_pydeseq2
• Add compute_all parameter to decide if applying DE analysis to all count matrices

Renam fgsea function to gsea, and modify:

• Add method parameter: default is "blitzgea"; alternatively support "fgsea"
• Set min_size to 5, and max_size to 4000
• For blitzGSEA, center by default is False

Compatible with matplotlib v3.9+:

• Change legend.legendHandles to legend.legend_handles, as v3.9+ removes legendHandles attribute.

Improve calc_dendrogram function:

• Simplify logic on rep by using update_rep and X_from_rep utility functions