(EAI-653): Run LLM-as-a-judge evals on all rated messages

[mongodben]

Jira: https://jira.mongodb.org/browse/EAI-653

Changes

• Run LLM-as-a-judge evals on all rated messages
• Standardize updateTracing func for all routes

Notes

• Note: there'll be a bit of re-running LLM-as-a-judge evals on rated messages where a message is already LLM as a judge evaluated. Given that approx 2% of messages are rated, and we're running evals on on 10% of all messages. This means that evals will be rerun on .2% of all messages (.1 * .02).

[nlarew]

LGTM

re: the re-run evals, doesn't seem like a big deal ultimately. That said, should we store info about evals in the conversations collection? e.g. a flag (e.g. Message.hasEval) or pointers to the evals in braintrust (e.g. Message.evals = ["<link to braintrust>"]?

[mongodben]

re: the re-run evals, doesn't seem like a big deal ultimately. That said, should we store info about evals in the conversations collection? e.g. a flag (e.g. Message.hasEval) or pointers to the evals in braintrust (e.g. Message.evals = ["<link to braintrust>"]?

thats an interesting idea but i'd rather not so that the tracing/online evals can be fully non-mutative, just exists on top of the existing behavior