Thank you, Louis, for the swift response. That helps a lot!
Indeed the query on databricks is <1 minute, while the refresh time in PowerBI is about 12 minutes. So it is the data transfer/bandwith which is causing the issue.
All points clear, just one follow-up question.
I was looking into your suggestion #1 (pruning): What exactly should I look for? I saw that "Files read" is a bit less than the distinct values of `a` in that query and the percentage is about 30%.