google_ai: Parse thought parts in Gemini responses

[dogonthehorizon]

Closes #31902. Though I'm not sure how completely. I noticed that function parts are being ignored so I did the same. Not sure if this impacts the ability to show thinking in the agent window, as I'm rather unfamiliar with what's happening behind the scenes. This small fix enables me to use Gemini 2.5 Pro Preview again.

Release Notes:

• Fixed thought parsing issue with Google Gemini 2.5 Pro Preview

[cla-bot[bot]]

We require contributors to sign our Contributor License Agreement, and we don't have @dogonthehorizon on file. You can sign our CLA at https://zed.dev/cla. Once you've signed, post a comment here that says '@cla-bot check'.

[dogonthehorizon]

@cla-bot check

[cla-bot[bot]]

The cla-bot has been summoned, and re-checked this pull request!

[lovasoa]

I think this should be merged since this is a strict improvement on the current state.

However, that would still not allow zed to re-use a previous chain of thought during a subsequent question in the same thread.

[osyvokon]

@dogonthehorizon, thanks for the fix!