Explicit Task and Model Definitions

[davidslater]

We should be significantly more explicit about what the model needs to do (the task), what the model is actually doing, and potentially what the attack threat model is.

• [ ] Tagging model output (output format - transcript or prob lattice...; logits vs probs vs log probs)
• [ ] Explicit task / scenario definitions, including preprocessing
• [ ] Threat model definition

[davidslater]

Updated to reflect changes since 0.12