Cuda versioning

[eitsupi]

• Remove cuda10.1 images (R 4.2.2 or later?)
• Switch cuda11.1.1 to cuda11.7.X (R 4.2.2 or later)

Related to #282

[eitsupi]

@cboettig It appears that the X.Y.Z-cudnn8-devel-ubuntu22.04 tag currently used in the base image does not exist for jammy.
Do you think 11.7.0-devel-ubuntu22.04 should be used as the base image instead?

rocker-versioned2/dockerfiles/cuda11_devel.Dockerfile

Line 1 in bfecf9e

FROM nvidia/cuda:11.1.1-cudnn8-devel-ubuntu20.04

https://hub.docker.com/r/nvidia/cuda/tags?page=1&name=22.04

(Edit)

It seems a issue https://gitlab.com/nvidia/container-images/cuda/-/issues/162.
It is expected to be released once cudnn supports cuda 11.7.

[eitsupi]

Due to a change in the stack file structure between 4.2.0 and 4.2.1 in #506, the automatic update of the stack file of the previous version was temporarily stopped (e4028c4).
Since the stack file structure change also occurs between 4.2.1 and 4.2.2 to drop the cuda10.1 images, I think we can keep this up for a while and re-enable automatic updates after the R 4.2.3? release.

[cboettig]

sounds good. Yes, I think we also want to drop cuda10.1 on the 4.2.2 and make the default ml image build on cuda 11.1