feat: Anthropic extended thinking and interleaved thinking

[WtXeLd]

Description:

The Python SDK recently added support for Anthropic's extended thinking and interleaved thinking capabilities (merged in this PR). This feature is currently missing in the JavaScript/TypeScript SDK.

Current Behavior:
The JS/TS SDK does not support Anthropic models' thinking blocks, which are required as the first content in assistant messages when reasoning is enabled.

Requested Feature:
Add support for Anthropic extended thinking in the JS/TS SDK, similar to the Python implementation:

• Store thinking text in response reasoning items
• Handle both streaming and non-streaming cases
• Support interleaved thinking when using tools with Claude models

Motivation:
This feature is critical for our project as we rely on Claude models' reasoning capabilities. Having parity between the Python and JS/TS SDKs would enable us to use the same advanced features across our stack.

Reference:

• Anthropic Extended Thinking Documentation
• Python SDK implementation for reference

[seratch]

Thanks for sharing this. The Python enhancement was a fantastic contribution by a community developer. If anyone is willing to apply the same to our ai-sdk adapter module in a good quality, we're happy to have the changes.