perf: enable zero-allocation queries and memory-mapped indexes

Introduce zero-allocation row extraction and query result buffers across `obicompactvec` and `obikpartitionner` to eliminate per-kmer heap allocations. Replace in-memory MPHF deserialization with memory-mapped, zero-copy views to reduce runtime memory footprint. Add configurable I/O chunking, a RAM-aware `--chunk-size` parameter, and system memory monitoring via the new `sysinfo` dependency. Re-export `PreloadedIndex` for external consumers.

src/obicompactvec/src/bitmatrix.rs

@@ -32,6 +32,14 @@ impl PersistentBitMatrix {
         self.cols.iter().map(|c| c.get(slot)).collect()
     }
 
+    /// Fill `buf[i]` with `col_i[slot]` as 0/1 u32, without allocating.
+    /// `buf` must have length ≥ `self.n_cols()`.
+    pub fn fill_row(&self, slot: usize, buf: &mut [u32]) {
+        for (c, col) in self.cols.iter().enumerate() {
+            buf[c] = col.get(slot) as u32;
+        }
+    }
+
     /// Returns the number of set bits in each column as `Array1<u64>`.
     pub fn count_ones(&self) -> Array1<u64> {
         let counts: Vec<u64> = (0..self.n_cols())

src/obicompactvec/src/intmatrix.rs

@@ -33,6 +33,14 @@ impl PersistentCompactIntMatrix {
         self.cols.iter().map(|c| c.get(slot)).collect()
     }
 
+    /// Fill `buf[i]` with `col_i[slot]`, without allocating.
+    /// `buf` must have length ≥ `self.n_cols()`.
+    pub fn fill_row(&self, slot: usize, buf: &mut [u32]) {
+        for (c, col) in self.cols.iter().enumerate() {
+            buf[c] = col.get(slot);
+        }
+    }
+
     // ── Distance matrices ─────────────────────────────────────────────────────
 
     pub fn bray_dist_matrix(&self) -> Array2<f64> {