[Bugfix] Fix the issue with interleaved thinking when using streaming

[chaunceyjiang]

Purpose

Fix the issue with interleaved thinking when using streaming.

｜begin▁of▁sentence｜>

## Tools
You have access to a set of tools you can use to answer the user's question.

<think>...thinking about results</think>

The current date is: 2024-09-30<｜User｜> What is the weather like in 
Beijing today? And tomorrow? <｜Assistant｜><think>

When the user's prompt contains something like <think>...</think> ... <think>, is_reasoning_end incorrectly determines that reasoning has already ended. In reality, because the last token is <think>, the reasoning has just begun. This causes reasoning to fail to work properly.

Especially in interleaved thinking mode, where the chat_template renders a <think>...</think> ... <think> prompt with 100% consistency.

Test Plan

Test Result

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan, such as providing test command.
• The test results, such as pasting the results comparison before and after, or e2e results
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
• (Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

[gemini-code-assist]

Code Review

This pull request addresses a bug in is_reasoning_end that caused incorrect behavior with interleaved thinking blocks during streaming. The change correctly determines if reasoning has ended by checking for the last occurrence of a start or end token. The logic is sound and effectively resolves the issue. However, the pull request lacks unit tests to cover the fixed scenario, which is a significant omission for preventing future regressions. I have added a comment to highlight the need for adding a dedicated test case.