Stop asserting token counts for multimodal chat completions

[alexander-alderman-webb]

OpenAI Chat Completions does not return token counts by default.
Text-based input and output can be estimated with tiktoken, but we consider estimating multimodal tokens out of scope.

[constantinius]

The test case was fixed to include the correct streaming option so that we could get the token counts. Additionally a PR in the Python SDK allows us to capture token usage in streaming completions.