Extract base_model_udf.load_models into separate class

[MarleneKress79789]

Background:

• Currently, we use from_pretrained with the cache_dir option to load the model and upload the cache directory to the BucketFS
• The current approach in general works, but seems to check each time online if the model exists, we have access and we have the newest version
• This is not possible in some deployments
• To solve this issue, we could follow the following example using save_pretrained which stores the model into a directory which we then could upload
• We can then load the model with from_pretrained from a local path
• It would be easier to change the model loading once it is extracted into its own class

Acceptance Criteria

• [ ] Extract base_model_udf.load_models into seperate class and inject it into base_model_udf