- Switch out implementation for generating parallel CJS build
- Follow guidelines from here microsoft/TypeScript#54567

- errors throw from sync and async functions should result in errored run
- errors thrown by promise of async generator should result in errored run containing all output until then
- interrupted promise of async generator should result in cancelled (errored) run containing all output until then
- regular async generator (without promise wrapper) should be supported in success, error and interrupt cases

…e traceable doesn't modify return value of wrapped function

- if function returns a subclass of promise return proxy object combining original return value and "traced" promise
- if function returns async generator (without promise wrapper) do not wrap it in a promise
- handle errors thrown in generator functions
- handle errors thrown in sync functions
- handle generator cancellations
- handle errors thrown by tracing client

- Use custom return value methods on openai[.chat].completions.create()
- Test that call-time metadata is attached to run

…onfigurable arg

- also fix return type of patched openai sdk (it was missing lots of methods on the original)