Also added some high-level tests to minimize the chances that we're breaking anything.
The typescript SDK is mostly functional at this point, with the exception that we don't have a build process or way to import it when deployed as an NPM package.
Ok so this is still pretty rough, and notably there's no reporting for streaming. But for non-streaming requests I've verified that this does in fact report requests locally.
Auth logic isn't shared between the clients anyway, so co-locating them is confusing since you can't use the same clients to call both. This also makes the codegen clients less verbose.
This makes it so we're using our own openpipe client for all OpenAI calls from the OpenPipe app.
The client doesn't do anything at the moment beyond proxying to the OpenAI lib. But this infra work should make it easier to quickly iterate on the client and test the changes in our own app.
