Dear Bastian,
welcome and thanks for the nice summary and explanations. I am using a similar setup and it works nicely with small collections. With larger collections the DT option " Storage/XMP: enable “look for updated xmp files on startup" will make DT unresponsive for quite some time. My strategy was to use two libraries in Darktable: A small “current” one and a larger “archive” one, that doesn’t look for XMP files on startup. However, I see my workflow changing because although collections and collection filter in DT have some issues (see other posts in the forum: A calm, rational and respectul discussion about the new collection filters), they are nevertheless quite powerful, and I am mainly using darktable for tagging etc. nowadays.