feat(aws-bedrock/anthropic.claude-sonnet-5): add new models [bot]

[models-bot]

Auto-generated by model-addition-agent for aws-bedrock/anthropic.claude-sonnet-5.

---

Note

Low Risk
Metadata-only addition with no runtime or application logic changes.

Overview
Adds AWS Bedrock catalog entry for anthropic.claude-sonnet-5 via new providers/aws-bedrock/anthropic.claude-sonnet-5.yaml.

The definition marks the model active, serverless, chat mode, with extended thinking enabled. It documents a 1M-token context window and 128k max output, text/image input, and standard Claude tool features (function calling, prompt caching, cache control, etc.). Per-region token pricing (including batch and cache read/write) is listed for ap-southeast-1 and ap-northeast-2 only, with links to AWS and Anthropic pricing docs.

^{Reviewed by Cursor Bugbot for commit 3c7dcdf. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 3c7dcdf. Configure here.}

[Archit-Kumar-16]

Already handled by the gateway without the flag. normalizeSystemRoleMessages ORs the feature against claudeVersionKeepsMidConversationSystem() (src/utils/modelParamUtils.ts:476-480), whose version table is:

MID_CONVERSATION_SYSTEM_FROM = { fable: [5,0], mythos: [5,0], opus: [4,8], sonnet: [5,0] }

Opus 5 and Sonnet 5 both clear that threshold, so mid-conversation role: system messages are preserved for these ids as written. Adding the explicit feature would be harmless but changes no behaviour.