Add enjim dataset(s)

[lloorree]

To run/set up:
- download sqlite from https://www.sqlite.org/download.html and install
- install/update dependencies however you're managing the .toml requirements locally
- update catbox paths in config.yaml
- run normally

To sanity-check the output (basically checks for bizarrely-formatted garbage):

grep -E "<[^* '\n3A-Z=<\.]+>|\[[^A-Z* 0-9r\.]{0,10}\]|&[^ \n]{1,10};|:[^ \n0-9]{1,10}:" rev-020c8d0-args-d93e21a.jsonl -c

[0x000011b]

Absolutely top notch work, thank you! Just a couple minor points to address.

[0x000011b]

I assume you were going to implement the registry pattern for the modules but backed away from doing that in this PR?

This import causes a crash since the file doesn't exist, but removing it and the reference below fixes it since it's not used elsewhere.

[0x000011b]

A couple minor issues with the personas:

• Empty personas are generated sometimes, resulting in processed text that looks like:

  A's Persona: 
  B's Persona: (actual persona text here)
  Scenario: (proper scenario here)
  ... 
  

  Which is a little wasteful token-wise and might be confusing to the model when training.
• They seem to be unparsed query results instead of cleaned text, so instead of generating

   A's Persona: A is like this and that
  

  They look like:

  A's Persona: ("A is like this and that", )
  

  This is also happening on world_scenario.