[CUDA] Initial implementation of Convolution with cuDNN (#2385)

* Link with cuDNN

* Initial implementation

* Remove backend apis

* Fix recording cudnn conv

* More unused backend apis

* Fix C++ conv tests

* include cudnn as python dep

* Install libcudnn9-dev-cuda-12 in CI

* cudnn only accepts contiguous inputs

* Switch to backend apis

* Plan needs to be kept alive

* Turn off tf32

* Add cache

* Test the native cuda graph api

* Set cudnn stream before execution

* Make LRUCache more like a normal container

* Do error check for cublas handle

* Zero-initilizing array

* Use tf32 for conv

* Skip TestConv.test_torch_conv_2D test

---------

Co-authored-by: Awni Hannun <awni@apple.com>

mlx/backend/cuda/utils.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <cublasLt.h>
 #include <cuda.h>
 #include <cuda_runtime.h>
 
@@ -33,10 +34,12 @@ class CudaStream {
 };
 
 // Throw exception if the cuda API does not succeed.
+void check_cublas_error(const char* name, cublasStatus_t err);
 void check_cuda_error(const char* name, cudaError_t err);
 void check_cuda_error(const char* name, CUresult err);
 
 // The macro version that prints the command that failed.
+#define CHECK_CUBLAS_ERROR(cmd) check_cublas_error(#cmd, (cmd))
 #define CHECK_CUDA_ERROR(cmd) check_cuda_error(#cmd, (cmd))
 
 // Convert Dtype to CUDA C++ types.