perf(server): generate chunk baselines outside the cache lock

crates/server/src/chunk_cache.rs

@@ -26,8 +26,12 @@ impl ChunkCache {
     }
 
     /// Returns the baseline for `pos`, generating and caching it on a miss.
+    ///
+    /// Generation runs outside the lock, so concurrent callers do not serialise on a cache miss. Two callers racing on the same position may each generate a baseline; generation is deterministic and side-effect-free, so the duplicated work is redundant rather than incorrect, and is far cheaper than serialising every miss behind the store.
     #[must_use]
     pub fn get_or_generate(&self, pos: ChunkPos, generator: &VoxelGenerator) -> Chunk {
+        // Probe under the lock, then release it before generating.
+        {
             // A poisoned lock cannot yield a usable cache; recovering the guard lets generation proceed rather than propagating a panic across every worker that shares this cache.
             let mut guard = self
                 .inner
@@ -36,8 +40,16 @@ impl ChunkCache {
             if let Some(hit) = guard.get(&pos) {
                 return hit.clone();
             }
+        }
+
         let chunk = generator.generate_chunk(pos);
-        guard.put(pos, chunk.clone());
+
+        // Retake the lock only to publish. A concurrent caller may have inserted the same position in the meantime; overwriting is harmless because both baselines are identical.
+        self.inner
+            .lock()
+            .unwrap_or_else(std::sync::PoisonError::into_inner)
+            .put(pos, chunk.clone());
+
         chunk
     }
 }