Module sparse_streaming 

Lock-free parallel streaming pass over a SparseIoVec that yields a merged SparseRunningStatistics<f32>.

Both HVG selection (hvg::select_hvg_streaming) and NB-Fisher weighting (gene_weighting::compute_nb_fisher_weights) need the same per-feature (npos, sum, sum_sq, n_cells) aggregates from a sparse backend. Sharing the skeleton keeps the parallel / lock-free invariants in one place and gives every caller a labelled progress bar with consistent styling.

Functions

streaming_sparse_running_stats

Stream cells through data_vec in block_size chunks, accumulating per-feature NB sufficient statistics into one SparseRunningStatistics per rayon worker, then merge to one result.

streaming_sparse_running_stats_folded

The same pass, accumulating a second set of statistics on a folded axis.