rapidsai
Change databricks container to multi-stage build
Currently, there is a frozen conda environment file that is used to install RAPIDS into the databricks base container.
This frozen file needs to be generated manually from the latest docker image and committed to the repo (thanks @skirui-source for doing this for 22.06 in #177).
In this PR I've updated the docker image to use a multi-stage build that exports the frozen env from the RAPIDS image, copies it to the databricks base and installs it there.
This should save some manual steps and mean that updating the image in the future will just mean changing the image tag in the Dockerfile instead of having to generate the whole frozen environment. It also means if the docker images get updated to fix security vulnerabilities, etc then next time someone builds the image here those will be picked up.
I've tweaked this further in 56466f7 and moved the RAPIDS image to a build arg. This means the only place we specify the image is in the README.md and hopefully makes it a little more clear to the user which version of RAPIDS they are building a container for. When we need to update this next time it means we only need to modify the README.
To me this is more clear, but if it's worse for others we can drop this commit and leave the RAPIDS image name within the Dockerfile.