[CUDA] Simplify allocator (#2392)

* simplify allocator and fixe race with small pool

* Don't use shared event in worker

* use cuda buffer in small pool

* comment

* comment

mlx/backend/cuda/eval.cpp

@@ -19,8 +19,6 @@ void new_stream(Stream s) {
   cudaFree(nullptr);
   // Ensure the static stream objects get created.
   cu::get_command_encoder(s);
-  // The main thread is safe to free buffers.
-  cu::allocator().register_this_thread();
 }
 
 void eval(array& arr) {