build libcuml 22.x

[yitao-li]

• Check https://github.com/rapidsai/cuml/releases for stable releases of libcuml
• Try adding 22.x to

  libcuml-builds/.github/workflows/cuml.yml

  Line 14 in e7c5b49

  cuml: ['21.08', '21.10']

  once libcuml 22.x is released. In the best case scenario, the build will succeed after about 2 hours [*]. If not, then you may need to make some (hopefully minor) 22.x-specific adjustments to the build process to make it work.
• Finally, make a PR in the {cuda.ml} repo to add the new libcuml version to https://github.com/mlverse/cuda.ml/blob/d76add4d93950b1c85356823afe00bd11867ff8d/.github/workflows/R-CMD-check.yaml#L18, and hope all existing features work with this new version of libcuml.

[*] As you may have noticed, the builds are now incremental. So, once you add a new version, only the new version of libcuml will be built because the GitHub action workflow sees there are already pre-built versions of libcuml 21.x that are released. Therefore to re-build and re-release an existing supported libcuml version (should it become necessary), you must delete the existing release and re-run this workflow manually. The "re-build and re-release" scenario seems highly unlikely though. I highly doubt we'll ever run into it.