kv-cache : avoid kv cells copies (#24277)

src/llama-kv-cache.h

@@ -269,7 +269,9 @@ private:
     // TODO: temporary until we refactor to be able to share the same cells between 2 kv caches [TAG_KV_CACHE_SHARE_CELLS]
     llama_kv_cache * other;
 
-    std::vector<llama_kv_cells> v_cells;
+    std::shared_ptr<llama_kv_cells_vec> v_cells_impl;
+
+    llama_kv_cells_vec & v_cells;
 
     // maps from a sequence id to a stream id
     std::vector<uint32_t> seq_to_stream;