Add alerting and/or SLOs to monitor the LLM plugin (for GA support)

[jtheory]

It doesn't seem like we're making breaking changes to the LLM app -- if we flip the label to GA, are we confident we'll keep it running & stable?

E.g. we should be alerted if the OpenAI connection drops, or users with the OpenAI connection configured are hitting a lot of failures in Cloud.

We probably don't want to be woken up if someone has configured a private LLM connection and are running into problems, since we have no control there; but maybe a warning would be useful.