ci: sets ubuntu versions to 26.04 and 24.04

[yosefmaru]

Updates Dockerfiles to use latest available Ubuntu (26.04 or 24.04) where applicable

Base image and dependency upgrades:

• Updated the CUDA and Ubuntu base images in cuda13-miniconda-jupyterlab/Dockerfile and lightning-jupyterlab-py39/Dockerfile to use CUDA 13.2.1 and Ubuntu 24.04, and added a new Dockerfile for a Miniconda-based JupyterLab environment
• Updated the RStudio Dockerfile (r2u-rstudio/Dockerfile) to use the newer rocker/r2u:26.04 base image, upgrading from Ubuntu 22.04 to 26.04.
• Updated the runtime base image in ubuntu-ffmpeg/Dockerfile from Ubuntu 24.04 to 26.04.

[galenlynch]

I think we should use miniforge instead of miniconda for cuda13-miniconda-jupyterlab/Dockerfile‎, since it's open source.

Does pytorch support CUDA 13.2 yet? I thought most wheels had 12.8 and 13.0 support. 13.2 appears to be experimental

We can use multi-stage docker builds instead of installing miniforge from scratch. The ubuntu-ffmpeg docker image is an example of a smaller build that doesn't require installing a bunch of apt packages to add conda.

I think nvidia/cuda:13.0.3-base-ubuntu24.04 is a more appropriate base image, because the base family is much smaller and pytorch often installs all the runtime libraries anyways, so having it in the base image isn't helpful for a pytorch-oriented image.