In the new year I might have to step back from all this and start a NEW script that takes into account the most logical order for this, so that we:

• store some sort of digest of the status of the post-clone-cleanup, so we have a state from which to determine "do I need to clone again?" beyond just the contents of source.json, which isn't sufficient anymore
• clone each repo or SPECIFIC repos (to avoid doing 40 clones if you only want 1) - my --force-clone isn't doing what I want, see previous point
• avoid re-cloning backstage if we don't have to, but can still resolve the backstage manifest.json even if we don't re-clone the repo, to resolve backstage: refs
• generate the manifest.json for each workspace so we can use that to remove workspace: refs
• perform the plugin exports, using the package-list.yaml to filter which ones we want to export
• keep the dist-dynamic content we need (embedded, package.json, yarn.lock), delete the rest
• delete examples/ and packages/ {app,backend}

  and plugins /* we didn't export (like -common)

• commit changes to the GL repo
• trigger rebuilds using the generatePipelineruns.sh --trigger command

That might end up being several scripts in a trenchcoat (so to speak) but it would make sync-midstream.sh much easier to read, if it's just calling out to other scripts

It would also help with single-workspace updates

Another thing to consider is that we might have to commit the output of `npx rhdh-cli plugin package` to a new repo

• do all the exports (including generating .oci files) in GL pipeline, non-hermetically;
• commit those to a new GL repo (better performance so the rhdh-plugin-catalog repo remains the SOURCE of truth, not binaries)

That might make the SBOM creation happier, but would as a consequence mean we're not really building from source IN Konflux, just fetching sources and copying previously-built stuff stored in GL into the scratch images as oci artifacts.