Optimize image sizes for `root` deployment

[dotsdl]

The image sizes for each of our root Docker builds come in at over 5 GB each, which is a bit surprising especially for the server image which doesn't require cudatoolkit (we use jaimergp/label/unsupported-cudatoolkit-shim to avoid pulling it in, but perhaps this no longer works).

If possible, we would very much like to shrink this down, at the very least to half this size, but I imagine it's possible to go lower.

[dotsdl]

@mikemhenry any thoughts on what might be ballooning our image sizes here?

[dotsdl]

For reference, doing a docker pull mambaorg/micromamba:1.5.10 to pull down our base image gives an image that's only 89.0MB in size. Our biggest image layer is 2.11GB when doing e.g. docker pull ghcr.io/openfreeenergy/alchemiscale.org-root-fah-compute:2024.11.21-0. Decompressed the overall image is then 5.23GB.