This is a very small change that would greatly facilitate writing CSVDataset interfaces for databases with multiple file extensions (cf. #160 (closed)).
I propose to simply strip the extension from the file path when creating the sample key, which enables to also use file paths containing extensions.
Nothing needs to be changed for the data loading : if using paths with extension, we can just set
extension="" when initializing the CSVToSampleLoaderBiometrics.