Add correct dependencies for mlir-nvidia dockerfile
This fixes some missing dependencies with the mlir-nvidia dockerfile.
The docker image can now be built and run to have an environment for building opencarp with MLIR CUDA code generation.
When building opencarp with MLIR CUDA in the container, I found that one must specify clang and clang++ as the C and CXX compilers, and
add the --cuda-feature=+ptx64
flag when compiling (-DCMAKE_CXX_FLAGS='--cuda-feature=+ptx64'
). This is because the CUDA version available
with the package manager in the container is a bit older than the default version clang assumes.