implement token_based_document_to_text_based()

[ArneBinder]

in analogy to text_based_document_to_token_based() (see #315).

This also adds the following improvements to text_based_document_to_token_based():

• use tokens, token_offset_mapping, and char_to_token from metadata, if available there and not explicitly provided, and warn, if provided, but different to entries in metadata
• raise an exception if the layers that target the text (the layers we want to modify) consist of annotations that are no Spans