Unify `method`/`flavor`/`backend`/… parameters

I’d classify the kinds of selections we make into

• recipe/algorithm/statistical test/…
• backend/implementation package/…

Of course there’s overlap, but I think that’s broadly it.

Currently, we have the following.

• flavor

  • sc.pp.highly_variable_genes: choose algorithm/recipe (seurat, seurat_v3, cell_ranger)
  • sc.tl.louvain: choose implementation package (vtraag’s louvain package, igraph, rapids)

• method

  • sc.pp.neighbors: choose algorithm to calculate connectivity (gauss or umap). kNN impl is transformer

  • sc.tl.umap: choose implementation package (umap or rapids)

    Note
    should probably be changed

  • sc.tl.rank_genes_groups: choose differential expression test (t-test, wilcoxon test, …) (also has corr_method for p-value correction)

  • sc.tl.marker_gene_overlap: choose how to determine overlap (overlap_count, overlap_coef, jaccard)

• backend: nowhere

• algorithm: nowhere

Start by comparing the public parameters in sc.pp.highly_variable_genes, sc.tl.louvain, sc.pp.neighbors, sc.tl.umap, sc.tl.rank_genes_groups, and sc.tl.marker_gene_overlap. Review how each parameter selects an algorithm, implementation package, or statistical method; done means the naming scheme is consistently defined and applied across the affected entry points.