feat: publish eic_cuda and eic_dev_cuda as nightly as well

To limit the number of container builds, we only deploy 'default' containers for CUDA: with the latest released versions. We won't deploy (or test) 'nightly' containers for CUDA.

This PR changes that and causes builds and deploys for nightly containers for CUDA as well. In concert with this change, I've asked @sly2j to bump the number of container builds from 4 to 6 so we can still run them all simultaneously.

I am also working on reducing the container build time by a factor of 5 (estimated), but that is regrettably not ready yet.