ARROW-13344, we should handle the case in which the user calls summarise() but does not call any aggregate functions in it.
Currently this is unhandled and results in errors (which are then are converted to warnings) like:
or when the expression in summarise() has only scalar literals:
or when there are zero arguments passed to summarise() and there is no group_by() preceding it:
We should aim for consistency with dplyr::summarise() in these cases, but if that's not practical then we should detect this case and throw a more informative error.