CUDA on Fedora 30

To install:
https://negativo17.org/nvidia-driver/

```bash
# Add repo
sudo dnf config-manager --add-repo=https://negativo17.org/repos/fedora-nvidia.repo

# Install nvidia driver
# ???

# Install all CUDA stuff
sudo dnf -y install nvidia-driver nvidia-driver-cuda cuda-devel
```

To get a compatible gcc:
https://devtalk.nvidia.com/default/topic/1052164/cuda-setup-and-installation/cuda-10-installation-on-fedora-30-with-gcc-9-1-1-20190503/

```bash
sudo dnf install cuda-gcc cuda-gcc-c++

# then link them up to cc and c++ in your PATH so `nvcc` will work without having to add a `-ccbin`
```

Also, from https://github.com/torch/torch7/issues/1138#issuecomment-381380180:

There is a bug in cmake's `FindCUDA` or something, the negativo repo installs CUDA to a weird place that cmake doesn't like, so you have to do

```bash
mkdir -p ~/cuda-dummy
cd ~/cuda-dummy
ln -s /usr/bin bin
ln -s /usr/lib64 lib
ln -s /usr/lib64 lib64
ln -s /usr/include/cuda include
ln -s /usr/share/cuda share
```