Display R1's thinking blocks in separate collapsable view in the UI

[joulei]

When talking with DeepSeek R1 models we get back a thinking block before the actual response from the assistant. It usually looks like this:

<think>
I'be been asked what is the ....
</think>
The capital of France is Paris.

Currently, since this is part of the response text from the assistant, the entire block of text is displayed as-is, including the tags.
We want to implement something similar to DeepSeek's UI:

Screen.Recording.2025-02-17.at.11.29.58.AM.mov

In the future we want to make this step replaceable or extendable, as search agents will display a list of the websites they're navigating, etc.
For now let's make the thinking blocks have it's own UI.