feat: implement NUMA-aware worker pools for merge command

Replaces the global Rayon pool with per-NUMA-node thread pools that pin worker threads to their respective nodes, leveraging Linux first-touch allocation to reduce cross-NUMA memory contention and improve cache locality. Integrates the `hwlocality` crate with a vendored build, includes graceful fallbacks for single-socket or non-Linux systems, and updates dependency constraints. Also adds installation and architecture documentation, and corrects parallelism detection in the partitioner.

src/obikpartitionner/src/merge_layer.rs

@@ -217,6 +217,9 @@ impl KmerPartition {
             }
         }
 
+        let n_src_layers = unitig_paths.len();
+        debug!("partition {i}: de Bruijn graph build start — {n_src_layers} source layer(s)");
+
         enum Pass1Data {
             File(PathBuf),
             Batch(Vec<CanonicalKmer>),
@@ -224,7 +227,9 @@ impl KmerPartition {
         }
 
         const BATCH: usize = 4096;
-        let n_workers = std::thread::available_parallelism().map_or(4, |n| n.get());
+        // Inside pool.install() this returns the per-NUMA pool size; outside
+        // it returns the global pool size. Both are the right value here.
+        let n_workers = rayon::current_num_threads().max(1);
         let capacity  = n_workers * 8;
 
         let dst_filter = Arc::clone(&dst_map);
@@ -311,18 +316,20 @@ impl KmerPartition {
             fs::create_dir_all(&new_layer_dir)?;
             let mut uw = Layer::<()>::unitig_writer(&new_layer_dir).map_err(olm_to_sk)?;
             debug!("partition {i}: unitig traversal start — {} nodes", g.len());
+            let mut n_unitigs = 0usize;
             g.try_for_each_unitig(|unitig| {
+                n_unitigs += 1;
                 uw.write(unitig)
             })?;
             debug!("partition {i}: unitig writer closing");
             uw.close()?;
-            debug!("partition {i}: unitig writer closed — dropping graph ({} nodes)", g.len());
-            let n = g.len();
+            let n_nodes = g.len();
+            debug!("partition {i}: unitig writer closed — dropping graph ({n_nodes} nodes)");
             drop(g);
-            debug!("partition {i}: graph dropped — starting MPHF build ({n} unitigs)");
+            debug!("partition {i}: graph dropped — starting MPHF build ({n_unitigs} unitigs)");
             Layer::<()>::build(&new_layer_dir, block_bits, evidence).map_err(olm_to_sk)?;
             debug!("partition {i}: MPHF build done");
-            n
+            n_nodes
         } else {
             drop(g);
             0