Normalized view stats

[fishidaho]

Make post-selection normalization explicit and reachable

What's wrong

A normalized view computes row_scale/gene_scale/col_mean once, at construction, over the whole array it wrapped
(NormalizedViewBase.__init__, src/vsparse/_norm_common.py). __getitem__ then applies those parent statistics to whatever sub-block is asked for. On a mixed population — two cell types differing in both read depth and marker-gene expression — indexing the view for one cell type differs from normalizing those cells by 52% relative
Frobenius norm:

view-then-index vs normalize(subset): rel Frobenius = 52.2%
index-then-normalize vs normalize(subset): rel Frobenius = 1.90e-15

Read as "the normalized data for these cells", this silently fits downstream analysis (parafac2, PCA, anything) to different data than the caller believes.

What this changes

Rather than redefine what indexing returns, this keeps both operations and gives them names:

• __getitem__ stays a window into this view's matrix — exactly toarray()[key] without materializing the full matrix — and now says so, including what it is not.
• select(rows, cols) is new: it renormalizes the selected sub-array on its own terms, equivalent to arr[sel].normalized(). It returns a view, not a dense block, so it still composes with @ and toarray().

select also fills a real gap rather than just wrapping something that already worked: selecting rows of a VCSCArray is a minor-axis selection, which falls out of _VCSBase.__getitem__ as a scipy array with no .normalized() at all. The raw-array route isn't uniformly available; select is.

Verification

tests/test_norm_selection.py (new, 22 cases across both array types):

• select() matches normalizing the selection directly to < 1e-12 relative error, on the mixed-population fixture;
• select() == the arr[sel].normalized() route where that route exists;
• select() returns a composable view — sub @ B matches the dense reference;
• column-only and both-axes selection (the two-index-array case selects a sub-block, not a broadcast diagonal — that was a real bug in the first draft, caught here);
• select() with no arguments == the whole view;
• __getitem__ == toarray()[key], pinning the window contract;
• the two disagree by > 10% on a realistic selection — so the distinction can't be quietly collapsed by a later change.