(EAI-506 redux): Refactor full response generation as a stateless function

[mongodben]

Jira: https://jira.mongodb.org/browse/EAI-506

Changes

• Refactor the response generation as a stateless function so that we can more easily evaluate outputs w/o a server
• Evaluate our whole RAG pipeline w/ braintrust autoevals
• Refactor autoevals to use azure openai instead of plain ol openai
• Upgrade typechat to "^0.1.0" b/c it was causing issues w/ braintrust CLI (see CLI Error: [ERROR] Could not resolve "readline/promises" braintrustdata/braintrust-sdk#356)
• Add Braintrust instrumentation to our server implementation to trace results in the evals.
  • Note: We can also easily trace w the braintrust observability module if ever choose to use that.
• Refactor server unit tests for simplicity

(Taken from #489)

Notes

• Revised version of (EAI-506): Refactor full response generation as stateless function #489. Reverted in Revert "(EAI-506): Refactor full response generation as stateless function" #497 due to issues caused in staging, see that PR description for docs on issues observed.