Remove tests which check that model output contains the right keys

[r-ash]

These fail and need to be updated in multiple places whenever we change the model config. See #310 (comment) for a discussion.

I wonder if instead we could

1. Add some "common" tests which run each of the language implementations, and run every model variant exported by each language. This would extend the existing "model-output-equality.yaml" GitHub action
2. Add a test in here which reads the config and checks that all "outputs" are exported from the model. That way we won't ever have to manually update it