[rcannood]

spearman_rank_sparse() assumed its input was a general CSC matrix, so calculate_similarity() and calculate_distance() silently returned wrong Spearman correlations for:

• a dgRMatrix (e.g. CSR from anndata): ranked along the wrong axis, also when called via dynutils:::spearman_rank_sparse() as in Fix CSR Spearman scores and pbmc_multiome failures, add a seed to all methods openproblems-bio/task_predict_modality#83
• symmetric input, including a dense matrix that happens to be symmetric (as(x, "CsparseMatrix") turns it into a dsCMatrix, which only stores one triangle)
• a dgTMatrix, which errored because it has no @p slot

Changes:

• .process_input_matrices() converts all inputs to a dgCMatrix whose columns are the vectors to compare. For a dgRMatrix with margin = 1 it reuses the slots as-is, so no transpose is needed.
• spearman_rank_sparse() uses the same conversion.
• Added tests comparing dense, data frame, CSC, CSR, triplet and symmetric inputs for both margins and all methods.

Runtime is unchanged.