[azopenai] Fix pipeline failures #25727
Description
Collection of issues I'm seeing in the current pipelines:
- Test failing because it's too exacting on token counts. There's no definition in the SDK about exactly how many tokens should be used. I have at test failing that assumes those values shouldn't change, but that's not contractual - it should be adjusted to just ensure that some tokens were used (ie: the struct is filled out).
- Pipeline code coverage dropped. We did have some issues with test recordings, and this might have caused us to drop some tests from the normal CI runs. Just want to see if we're missing anything critical there (we do have a lot of redundant testing), and if not, we can close off this issue.