CLI: bench sparse command

[jc-macdonald]

Summary

Implement vbpca-py bench sparse — benchmark sparse/masked VBPCA fit. Replaces scripts/benchmark_sparse_mask_explicit.py.

Depends on #40 (CLI scaffolding).

Usage

vbpca-py bench sparse
vbpca-py bench sparse --n 100,500 --p 100,500 --k 5 --miss 0.1,0.3,0.5 --reps 3 --output sparse_bench.json

Flags

Same as bench shape plus:

Flag	Default	Notes
--miss LIST	0.1,0.3,0.5	Comma-separated missingness fractions
--pattern	mcar	Missingness pattern: mcar, mnar_censored, blockwise, monotone

Output format

Same JSON structure as bench shape, with additional missingness and pattern fields per result entry.

Implementation notes

• Port from scripts/benchmark_sparse_mask_explicit.py
• Use the same mask-generation utilities as the stability analysis
• Include explicit-mask size statistics (nnz, density) in the output

Acceptance criteria

• vbpca-py bench sparse prints a table to stdout
• JSON output includes missingness fraction and pattern
• Multiple missingness levels work
• Sparse kernel actually exercises the C++ sparse path