Update Huggingface Model for URL Classification

[maxachis]

The current Huggingface Model does not make full use of the data it has available, and should be modified.

Currently, the /url-relevance model exists as a text classification model, which, per the documentation, relies on a single text input. This fails to take into account the various forms of HTML data that we have available.

For a new model, we would want to make use of the various HTML data we have, both in the training and in the deployment of the pipeline.