Add support for Claude Sonnet 4 on Bedrock

After Add Claude Sonnet 4 on Bedrock to Evaluation Ru... (#550188), we should make documentation changes to announce that Claude Sonnet 4 on Bedrock is now supported for Self-Hosted Duo.

Use Cases for Evaluation

• Chat
• Code Suggestions

Definition of Done

• Each model can be used to support the feature on all supported platforms
• Examine individual inputs and outputs that scored poorly (1-2 scores); Look for and document any patterns of either poor feature performance or poor LLM judge callibration. Iterate on the model prompt to eradicate patterns of poor performance.
• Achieve less than 20% poor answers (defined as 1s and 2s from an LLM judge, or less than 0.8 cosine similarity) using each supported model for those areas in which we do have supporting validation datasets.
• Quality results, based on LLM Judge scores 1-4 and/or cosine similarity are recorded in this issue's comments as distributions. For LLM Judges this means buckets of 1s, 2s, 3s, 4s. For Cosine similarity scores, this means buckets of similarity scores 0.9 and above, 0.8-0.89, 0.7-0.79 and so on.
• The traffic light system for self-hosted models has been updated to include scores, and the documentation has been updated to reflect any changes