[mlr3automl] Error there is no package called ‘mlr3learners’ in CI

[PGijsbers]

See e.g.: https://github.com/openml/automlbenchmark/runs/3499152666#step:7:18384

Looks like it has to do with Github's policy to only use PATs , though I am not clear as to why because all downloaded repos are public, not private.

[PGijsbers]

I think it should be fixed in an upcoming release, as per this open issue.

[PGijsbers]

So it will never be truly fixed, as the default PAT will still be rate limited. We'd need to create a way to add a custom PAT. The primary reason for this is CI, where the framework gets installed multiple times - I don't think we have a guarantee that a custom PAT will also not be affected by the rate limit. For large scale experiments it is advised to build an image with the preinstalled setup which bypasses this problem.