Complements embedding analysis functionality

[p-ferreira]

• adds embedding_config.yml mapping a "real" CustomRewardModel
• adds embedding logic to config + code
• adds config param in RandomGatingModel so it doesn't break (note about this on loaders/model.py line 25)
• changes analysis template to support embedding_plot config
• adds extra dependencies in requirements.txt

Note: Since the reward model is loading a GPT-J, the query part will take a long time to be executed. In order to perform a fast test/execution, comment the line run_query() in the main.py file.

[steffencruz]

Looks good!