Improments in generator to improve generalization
This MR includes multiple fixes to get the generator back working more reliably. I have tested this locally against alpine main and community repos with a single architecture and it seems to have worked pretty well. However, as the goal is to have this stable, I would like to set up my own mirror and this running for at least a week to prove its reliability. That is the main reason why this is marked as Draft.
Most relevant for this MR is that I have updated the repository to be the real upstream. The previously checked-out branch has two commits that have not made it upstream. The first one seems totally unused, so it should be safe. However, the second one modifies the behavior to change the output files naming, and that is reflected in the appstream data published by alpine. When using upstream it is data/edge/community/Components-$arch.xml.gz, while using the fork it becomes data/edge/community/Components-community-$arch.xml.gz. Is that actually needed? If it is, should we try to explain the rationale and send it upstream? Otherwise, I believe the repository change should not be a problem.
Ideally, we should be able to get appstream-generator from testing to community and then install it directly from the repos after 3.16. Would really simplify the Dockerfile :)