tie together the various constants used for parsing the redcap CSV

[alistairewj]

I refactored the plethora of hard-coded dictionaries we had into a single Enum with an Instrument class. The goal of this was to (1) tie together all these constants so we wouldn't forget to add a columns variable for a questionnaire with the name variable, (2) reduce the number of hard-coded values, and (3) reduce the complexity of multiple dictionaries interacting.

• The Instrument class contains the hard-coded constants we need for each type of redcap instrument, e.g. the session_id which unique identifiers it, the name of the JSON file with the columns, and the name of the FHIR schema that was automatically converted from ReproSchema
• I fixed a few typos in the name of the questionnaires
• I now impute the redcap_instrument column in the loaded CSV (so I renamed this to be specific to loading a redcap CSV). This makes it easier for the later code to filter rows/columns together, since we can rely on the redcap_instrument column always having a value
• Cleaned up the interface to the dataset class a bit. Now the find_* methods return a list of Path objects, no more awkward dictionary stuff. This also fixed a bug where multiple sessions for a participant were merged together implicitly. It's a better signature in general IMO.
• Removed a lot of the now redundant list_* methods
• Updated the tutorial and added some code on loading audio

[ibevers]

This look a lot better! I'm glad that this also led to some bugs getting fixed