Apply per-gene multiple testing correction for site selection metrics

[Copilot]

Site selection p-values must be corrected per gene (not across the full panel). The site comparison output should include both raw and adjusted p-values for downstream consumers.

• Per-gene multiple testing
  • Apply Benjamini–Hochberg correction within each GENE group in omega_comparison_per_site.py.
• Output surface
  • Emit p_value_adj alongside p_value for each site/AA grouping.

result["p_value_adj"] = np.nan
for gene, gene_df in result.groupby("GENE", dropna=False):
    valid_mask = gene_df["p_value"].notna()
    if not valid_mask.any():
        continue
    adjusted = benjamini_hochberg(gene_df.loc[valid_mask, "p_value"].to_numpy())
    result.loc[gene_df.index[valid_mask], "p_value_adj"] = adjusted