fix: dropped generation-prompt text in JinjaTemplateChatWrapper

[JoshuaWise]

Description of change

JinjaTemplateChatWrapper was discarding text in the template's generation prompt that comes after the assistant role header. Any static text the template emits between the role header and the model's content -- most importantly a pre-filled reasoning tag like <think> was silently dropped. This PR fixes that behavior, enabling many templates that rely on pre-filling a <think> tag to force the model into reasoning mode, or <think></think> to suppress it.

Fixes #634

Pull-Request Checklist

• Code is up-to-date with the master branch
• npm run format to apply eslint formatting
• npm run test passes with this change
• This pull request links relevant issues as Fixes #0000
• There are new or updated unit tests validating the change
• Documentation has been updated to reflect this change
• The new commits and pull request title follow conventions explained in pull request guidelines (PRs that do not follow this convention will not be merged)

[giladgd]

Thanks for the PR!
I noticed it just now after I pushed a fix for the issue you opened (#636)
I'll take a look at it now and see whether your solution is better

[giladgd]

We need to make sure that we don't force-open a thought segment on every render, we should only do it when the model response is empty because we're just starting to generate it.
The rendering also happens after function calls are done (when the results are being put in the context), after context shifts, and some other edge cases, so we shouldn't break those

[JoshuaWise]

Feel free to close this in favor of #636

[JoshuaWise]

I just realized, my fix here doesn't actually work for what it's intended to solve. I'm trying to get Qwen3.6-27B-Fable-Fusion working, and although the tests pass, it's not working in practice. Your PR #636 seems to be the more robust fix. Closing this.