Convert: generic HDF5 matrix and delimited text files #25

Part of #20. Depends on #21. Covers the remaining scanpy/anndata readers: read_hdf, read_csv, read_text.

Generic HDF5 (read_hdf)

scanpy's read_hdf(file, key) treats one HDF5 dataset as a dense X.

adata convert m.h5 -o m.h5ad --key /path/to/matrix \
    [--obs-names-key /path] [--var-names-key /path] [--transpose] [--sparse]

With no --key, list the candidate 2-D datasets (reuse adata ls logic) and exit.

Delimited text (read_csv, read_text, .tsv/.txt/.tab, optionally gzipped)

Dense matrix, first row = column names, first column = row names.

• Parse row blocks as a stream; never load the file.
• --sparse writes CSR via sparsify from core/convert.py (recommended for count tables).
• --delimiter, --first-column-names/--no-first-column-names, --transpose (many tables are genes × cells).
• Needs a tiny CSV reader without pandas — share it with Convert: Parse Biosciences split-pipe outputs #30.

Out of scope

Excel, .soft.gz, UMI-tools — listed in #31.

Tests

Compare with scanpy.read_hdf / read_csv / read_text; perf guard on row-block parsing.