Allow using Anthropic adapter for inline stratey

[sasugo]

I was debugging a little bit and found out that the inline_output method doesn't contemplate the object structure return by Anthropic, where the response text comes in the thinking index and not the text as expected by the method. So the result of this is a success but the output is undefined

[sasugo]

Would be nice to have in the doc somewhere that the extendend_thinking['default'] should be false so it can work wiht the inline strategy, until you have the feature implemented of course

[olimorris]

Appreciate you flagging this. I would very much welcome a PR to add this to the Anthropic adapter. We could detect whether self.opts.can_reason is true and work with the output in that way.