Automate testing of all models via API

[suurinie]

We should fetch the list of predictor ID's and run the API test requests for these, because then

• None of the models is forgotten by accident
• there will be no typos in identifiers.

There is already a route for this in the API: /health/ready/?model=[model]
and /healthz/ that for-loops over models.

All I really have to do is add proper exceptions, and make the end points return something more sensible than generic error.