Python: MistralAIChatCompletion sets metadata['usage'] to a tuple, so usage is missing from telemetry #14617

Describe the bug

In the Python MistralAIChatCompletion connector, metadata["usage"] is set to a one-element tuple instead of a CompletionUsage, because of a trailing comma (mistral_ai_chat_completion.py#L254-L259):

metadata["usage"] = (
    CompletionUsage(
        prompt_tokens=response.usage.prompt_tokens,
        completion_tokens=response.usage.completion_tokens,
    ),
)

As a result:

• the telemetry decorator only records token usage when isinstance(usage, CompletionUsage) (decorators.py#L422-L427), so Mistral chat spans carry no gen_ai.usage.input_tokens / gen_ai.usage.output_tokens;
• user code that reads result.metadata["usage"].prompt_tokens (as it does for the other connectors) raises AttributeError: 'tuple' object has no attribute 'prompt_tokens'.

Expected behavior

metadata["usage"] is a CompletionUsage, as in the OpenAI, Google, Bedrock and Ollama connectors.

Fix

Remove the trailing comma (the outer parentheses then become a plain grouping).

Platform

• Language: Python, semantic-kernel main (7a92cd4)

Found by reading the code while checking how connectors report usage to telemetry; it was not run against the Mistral API here. This report was drafted with Claude Code.