[CUDA][Integrate] Switch CUDATarget to ptx_kernel cc. (#19925)

This drops the local LLVM revert:
llvm/llvm-project@de7438e

Signed-off-by: hanhanW <[email protected]>

Author: hanhanW

compiler/plugins/target/CUDA/CUDATarget.cpp

@@ -545,7 +545,7 @@ class CUDATargetBackend final : public TargetBackend {
         };
 
         // Mark the entry point as a kernel.
-        setMetadataValueI32("kernel", 1);
+        llvmFunc->setCallingConv(llvm::CallingConv::PTX_Kernel);
 
         // Set the maximum number of threads in the thread block (CTA).
         auto exportOp = exportOpMap[funcOp.getName()];