[DOC] Update the Installation Guide to use runtime instead of devel images for Docker users using pip

In the RAPIDS Installation guide, under the CUDA Support Notes section, the recommendation is to use the devel flavour of the nvidia/cuda images rather than the runtime or base images in order for Numba to function properly.

This information is outdated, as all the CUDA wheels currently include most of the headers required by NVRTC for Numba to function properly.

To test this, I ran the vector addition example from the Numba for CUDA documentation, and was able to successfully run the code on both the base and runtime images with cuDF installed through pip on top of these Docker images.

The base image still does not support cupy without other libraries being loaded before cupy (although this should be supported soon), so we can update the recommendation to use the runtime image without any issues.

CC: @ncclementi @vyasr

Open the RAPIDS Installation guide and find the CUDA Support Notes section. Update the Docker recommendation for pip-installed CUDA wheels from the devel image to the runtime image, retain the stated base-image limitation, and preview or build the documentation to confirm the wording and links render correctly.