Convert: Parse Biosciences split-pipe outputs #30

Part of #20. Depends on the streaming MTX reader in #24.

Format

split-pipe writes DGE_filtered/ and DGE_unfiltered/, for the combined run and for each sample:

• count_matrix.mtx — cells × genes (unlike 10x)
• all_genes.csv — gene_id,gene_name,genome
• cell_metadata.csv — bc_wells,sample,species,gene_count,tscp_count,mread_count,…

Mapping

• X ← count_matrix.mtx through the Convert: MTX and 10x MTX folders, import and export (streaming) #24 reader, no transpose
• obs ← cell_metadata.csv, bc_wells as obs_names; sample, species as categoricals
• var ← all_genes.csv; var_names via --var-names gene_symbols|gene_ids

Options

• Accept either a DGE_* folder or the whole split-pipe output directory (find DGE_filtered automatically; --unfiltered to switch)
• --sample NAME selects a per-sample subfolder; --all-samples writes one output per sample
• Handle mixed-species runs (genome column; optional --species filter)

To check

• Confirm the layout on current split-pipe output.
• Recent split-pipe versions (≥ 1.1) may already write an .h5ad — if so, document it and focus this converter on older runs and the per-sample folders.

A minimal CSV reader without pandas is needed; share it with #25.

Tests

Small fixture folder; compare with the scanpy.read_mtx + pandas recipe from Parse's docs.