perf: optimize entropy computation by pre-filtering monomorphic minorants

Shift monomorphism filtering from the entropy scan layer to a lightweight, annex-only pre-pass. By replacing `Selection::All` with a pre-filtered subset, expensive per-genome resolution is strictly limited to non-monomorphic families. This avoids processing ~98% of minorants that are known to be monomorphic, while preserving positional read speedups for subsequent runs. The change is an internal performance refinement with no public API modifications.

src/obikphylo/src/siblings/entropy.rs

@@ -182,9 +182,17 @@ impl ShannonEntropyExt for KmerIndex {
 /// [`super::subsample::compute_selections`] whenever entropy-biased
 /// selection is requested; see `DevDocMD/architecture/siblings.md`,
 /// "Entropy-biased selection", for why this is the one place that pays a
-/// full, unsampled scan (every non-monomorphic minorant's `genome_mask`
-/// must be resolved to know its entropy) — every later read of the
-/// resulting file is a plain positional mmap lookup, not a re-scan.
+/// scan (every non-monomorphic minorant's `genome_mask` must be resolved
+/// to know its entropy) — every later read of the resulting file is a
+/// plain positional mmap lookup, not a re-scan.
+///
+/// Bounded to non-monomorphic minorants only
+/// ([`super::subsample::non_monomorphic_selection_layer`], a cheap
+/// annex-only pass — no cross-partition resolution), not
+/// `Selection::All`: monomorphism (`family_size() < 2`) is already known
+/// from the annex bits alone, so there is no reason to pay
+/// `scan_layer_families`'s expensive per-genome resolution for the ~98%
+/// of minorants that are monomorphic only to discard the result.
 pub(super) fn ensure_entropy_annexes(index: &KmerIndex, layer_dirs: &[PathBuf]) -> OKIResult<()> {
     let missing: Vec<usize> = layer_dirs.iter().enumerate()
         .filter(|(_, dir)| !dir.join(ENTROPY_ANNEX_FILE_NAME).exists())
@@ -215,10 +223,9 @@ pub(super) fn ensure_entropy_annexes(index: &KmerIndex, layer_dirs: &[PathBuf])
         let layer_dir = &layer_dirs[li];
         let mut builder = EntropyAnnexBuilder::new(minorant_counts[li] as usize, &layer_dir.join(ENTROPY_ANNEX_FILE_NAME))
             .map_err(OKIError::Io)?;
-        scan_layer_families(layer_dir, n_parts, n_genomes, with_counts, k, &cache, &Selection::All, |family_idx, mask, genome_mask| {
-            if mask.family_size() < 2 {
-                return; // monomorphic minorant — leave the sentinel, never sampled
-            }
+        let eligible = super::subsample::non_monomorphic_selection_layer(layer_dir)?;
+        scan_layer_families(layer_dir, n_parts, n_genomes, with_counts, k, &cache, &Selection::Some(&eligible), |family_idx, mask, genome_mask| {
+            debug_assert!(mask.family_size() >= 2, "Selection::Some(eligible) must only visit non-monomorphic minorants");
             if let Some((h15, _)) = family_entropy(genome_mask) {
                 builder.set(family_idx, h15 as f32);
             }

src/obikphylo/src/siblings/subsample.rs

@@ -121,6 +121,35 @@ fn reservoir_sample_layer(layer_dir: &Path, n_layer: usize) -> OKIResult<HashSet
     Ok(reservoir.into_iter().collect())
 }
 
+/// Every non-monomorphic minorant's `family_idx`, for one layer — not a
+/// sample, the *complete* set. Structural, annex-only, no cross-partition
+/// resolution (same discipline as [`reservoir_sample_layer`], just keeping
+/// everything instead of a bounded reservoir). Used to bound
+/// [`super::entropy::ensure_entropy_annexes`]'s `scan_layer_families` call
+/// to a `Selection::Some` of exactly the families that need their entropy
+/// computed, instead of `Selection::All` — monomorphic minorants (~98% of
+/// them, measured) are known to be ineligible straight from the annex bits
+/// already, with no need to pay `scan_layer_families`'s expensive
+/// per-genome resolution (`fill_sub_matrix_carries`) only to discard the
+/// result once `family_size() < 2` is checked *after* resolving.
+pub(super) fn non_monomorphic_selection_layer(layer_dir: &Path) -> OKIResult<HashSet<usize>> {
+    let annex = SiblingAnnex::open(&layer_dir.join(ANNEX_FILE_NAME))?;
+    let mut selected = HashSet::new();
+    let mut family_idx: usize = 0;
+    for slot in 0..annex.len() {
+        let Some(mask) = annex.get(slot) else { continue };
+        if !mask.is_minorant() {
+            continue; // doesn't advance `family_idx` either
+        }
+        let this_family_idx = family_idx;
+        family_idx += 1;
+        if is_non_monomorphic_minorant(mask) {
+            selected.insert(this_family_idx);
+        }
+    }
+    Ok(selected)
+}
+
 /// Total minorant count per layer (`is_minorant()`, monomorphic included)
 /// — sizes the entropy annex (`entropy_annex.rs`), which needs exactly one
 /// entry per minorant to keep its `family_idx` numbering aligned with