Store the output of AI tools

Currently we only store in the evals the input and outputs of the LLMs but the Agents talk to "Tools", bits of code or functions on the Drupal site that can return messages. Some of these messages are quite advanced (Such as the output of Search or a RAG search). Some are simpler (Adding a taxonomy term with id 1 and label "cheese").

We should store these in the evaluations too. They are passed into the assistant message so will usually be in that history but not necessarily always especially in the Agent history.