On a related note, there are Roslyn analysers about that try to detect these sorts of issues in C# code for at least MEL and Serilog, so maybe that's something useful to port? (not that I could say offhand how much effort there is in porting the message template code)

The converse ("you passed a function as a logging arg") is one of the G-Research analysers; seems sensible to have this one too!