feat(live): Handle input transcription differently for Gemini Live 3.1 models

[wuliang229]

For Gemini 3.1 Flash Live, input transcriptions are now marked as finished in the first response containing text, rather than being accumulated as partial updates. This aligns with the expected behavior of the 3.1 API.

[adk-bot]

Response from ADK Triaging Agent

Hello @wuliang229, thank you for contributing to the google/adk-python repository with this pull request!

We noticed that this PR introduces a feature/fix for Gemini 3.1 Live models. To help the review process proceed smoothly, please make sure your PR follows our contribution guidelines. Could you please update the PR description with the following:

1. Testing Plan: Please include a testing plan section describing how you tested this change (e.g., manual runner logs or unit test validation).
2. Logs or Screenshots: If possible, please share a log output or screenshot demonstrating the behavior of the input transcription fix with Gemini 3.1.
3. Associated Issue: Please link an associated issue if one exists, or format the description to outline the expected and actual behavior following our issue templates.

Your help in providing this context is highly appreciated and will save review time. Thank you!