Music rating

[shahbuland]

Add music rating mode to CHEESE as an example for audio

• Looks at directories populated by WAV files
• Allows labellers to score WAV files and write comments on them

Also some other QOL changes to general setup and text captioning

• Further abstractions to pipeline so that even less new code needs to be written for new tasks
• Abstractions to client so less new code needed there
• Update to text captioning gradio client to automatically fetch new data whenever user submits (i.e. no period where they are being shown empty data like before)

[shahbuland]

Summary of new changes:

• Further abstracted any music task to WavFolderPipeline
• An arbitrary music/audio related task can be setup with WavFolderPipeline easily
• Created an additional abstraction for any Pipelines that write to a huggingface datasets Dataset
• Added docstrings to code that did not have it previously as well as to new code

Additional changes that may be desired on top of this would likely be better suited as separate features in a different branch off main:

• Simplifying construction of new pipelines to a single function, so that everything needed for a new task can be made in a few lines by just calling some functions, rather than defining custom objects
• A full readthedocs style documentation
• Adding model to the loop with architext