Clean up the model list (detection vs classification, versions, dates)

Nachet presents a large list of models, each with a short description (for example
"Use a Swin transformer to classify the seeds"), including some local testing models.
Two problems:

- It is unclear whether each model does detection, classification, or both. Nachet
Mini splits this into two selectors: one detection model and one classification model.
- Every model has an empty version string and the same date, so there is no way to
tell which was trained most recently.

Port Mini's split detection / classification selection, or clean up the current list
so roles, versions, and dates are legible.

- [ ] Each model's role (detection, classification, or both) is clear, ideally via split selectors
- [ ] Version strings are populated
- [ ] Training dates are accurate per model rather than identical
- [ ] Local testing models are clearly marked