Apply Benjamini-Hochberg FDR to group correlation stars

[urme-b]

The group correlation cell starred 36 pairwise Pearson tests on n=10 with no multiplicity correction. Replaced the raw-p stars with Benjamini-Hochberg FDR q-values over the unique off-diagonal pairs; the diagonal is no longer starred and the title states n and the correction. On the committed data this drops the 'significant' count from 6 to 2 — a more honest picture at this sample size. No new dependency (numpy/scipy only).

Closes #6

[urme-b]

Superseded: #5 landed its own Benjamini-Hochberg cell (mms.stats.corr_matrix_fdr) on main, so the inline-FDR approach here conflicts and is redundant. The remaining real issue — the old uncorrected-stars cell left behind by #5 — is fixed by removing that cell in the replacement PR.