v1.2.0: Sparse matrix support

What's New

Sparse matrix support for guidedPLS()

X1 and X2 now accept Matrix::dgCMatrix (sparse matrices) in addition to base R matrices. This is particularly useful for single-cell data (scRNA-seq, scATAC-seq) where feature matrices are large and sparse.

Key changes:

• Internally uses the identity X_c Y = X Y - mu(1^T Y) to avoid materializing the dense centered matrix
• Only the small k × p cross-product matrix is dense; the input X stays sparse throughout
• SUMCOR mode (sumcor=TRUE) auto-converts to dense with a verbose message (p × p covariance matrices are inherently dense)
• Added Matrix to Imports

Performance (benchmark):

Scenario	Dense	Sparse	Speedup	Memory
3000 × 10000, 5% density	5.93s	2.29s	2.6×	240 MB → 18 MB (13× reduction)
5000 × 50000, 1% density (scATAC-like)	OOM	55.96s	—	2000 MB → 30 MB (67× reduction)

Other changes:

• Added comprehensive sparse/dense equivalence tests (31 expectations, tolerance 1e-8)
• Added CLAUDE.md for development guidance
• Registered missing test files in tests/testthat.R

Full Changelog: v1.0.0...v1.2.0