feat: add Gemini 3 Pro thought signature support

[acodercat]

Add Gemini 3 Pro thought signature support

Gemini 3 Pro enforces stricter validation on thought signatures for multi-turn function calling. This PR adds support for preserving and returning thought signatures to maintain reasoning context across conversation turns.

Without this fix, requests to Gemini 3 Pro return a 400 error when required thought signatures are missing.

References:

• https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures
• https://docs.litellm.ai/docs/providers/gemini#thought-signatures

[victorb-sierra]

Thanks for this PR. Not a bad idea to use the raw_data directly. Only worry is that since message object fields are mutable we could end up with a discrepancy between message content and raw content. We can make the behavior more explicit by forcing the choice of using raw_data or not

[acodercat]

Thanks for this PR. Not a bad idea to use the raw_data directly. Only worry is that since message object fields are mutable we could end up with a discrepancy between message content and raw content. We can make the behavior more explicit by forcing the choice of using raw_data or not

Thinks. It wroks well.