Consider refactoring `validate()`

[danielhuppmann]

@phackstock suggests to refactor the IamDataFrame.validate() function to only every return a pandas dataframe. If all data points satisfy the criteria, then an empty data frame should be returned instead of None. Dealing with two different return types usually makes for awkward code downstream and I'd consider it an anti-pattern.
If validate() would always return a pandas data frame, you would do something more readable like:

if not_valid.empty:
    # do something if everything's valid

or

if not not_valid.empty:
    # do something if there are some invalid data points

Originally posted by @phackstock in #930 (comment)