provides better perf with large recordings as it only loads the recording file once (but still allows splitting recording files for different endpoints)
currently only allows YAML persistence (VCR.py also allowed JSON), but should be easy to add if required
adds support for providing custom request forwarders (defaults to Azure OpenAI)
Add Document Intelligence example for forwarder config