Add Cohere Command as an LLM backend

[malexw]

This PR demonstrates adding Cohere's Command as a backend LLM for OpenGPTs. The code uses the XML agent as a starting point and makes some adjustments to the prompt, as well as adding a new method to stringify the tool schema in a way that Command responds well to.

One known issue is that even though Command does seem to do a good job of calling functions when appropriate, it still includes a normal text response after the function call. I'll trim that out from the log in the future, but I've left it in for now for testing and debugging.

[giannis2two]

Thanks for opening up the PR @malexw.

@nfcampos do you think the PR implementation is an acceptable design? Have noticed in the repo that similar LLM additions were added under backend/packages/gizmo-agent/gizmo_agent/agent_types/xml/ vs PR Cohere support is added under backend/packages/gizmo-agent/gizmo_agent/agent_types/. Thanks in advance.