It looks like Azure was having the same issue, it might just be all 12.2 Photon clusters.
https://community.databricks.com/t5/data-engineering/sqldf-bugs-in-gcp-workspaces/td-p/38578 
That post says it was been fixed, but last I checked it was still failing. 
Fortunately for me, I don't use it much - I just converted my SQL cells to spark.sql("") in a Python cell.