Correct unsupervised version

When no labels are provided, it is possible to use the unsupervised loss. The implemented version is not correct:
https://github.com/sylvchev/riemannianPCA/blob/277e4b9c91e61ba3ebe89e0cef19960698954757/riemannianpca.py#L142

Intermediate possibility is to use only one label for all the data.