Support Claude Sonnet extended thinking models

[avidspartan1]

Feature request

Add first-class support for Claude Sonnet extended thinking models in PR-Agent, including the current Claude Sonnet 4.6 model IDs across supported LiteLLM provider prefixes.

Problem

PR-Agent already has an enable_claude_extended_thinking configuration path, but model eligibility is controlled by a built-in allowlist. Newer Claude Sonnet extended-thinking models and provider-specific aliases are not recognized, so enabling the feature may not send the extended-thinking payload for valid Claude Sonnet deployments.

This affects users running Claude Sonnet through different LiteLLM routes, including Anthropic direct, Vertex AI, and AWS Bedrock regional/global model identifiers.

Requested behavior

• Recognize Claude Sonnet 4.6 extended-thinking model IDs in the built-in Claude extended-thinking allowlist.
• Include common LiteLLM provider forms such as:
  • anthropic/claude-sonnet-4-6
  • claude-sonnet-4-6
  • vertex_ai/claude-sonnet-4-6
  • bedrock/anthropic.claude-sonnet-4-6
  • regional Bedrock variants such as us, eu, au, jp, and global
• Provide a configuration override so deployments can replace the built-in extended-thinking model list without waiting for a PR-Agent release when providers add or rename model IDs.

Acceptance criteria

• When enable_claude_extended_thinking = true, supported Claude Sonnet 4.6 model IDs receive the extended-thinking request payload.
• Existing Claude 3.7 Sonnet extended-thinking behavior is preserved.
• A documented config key allows operators to replace the built-in Claude extended-thinking model list.
• Leaving the override empty continues to use the built-in defaults.

Reference

This request is based on local commit 98aab08a575ffeff4c642aea7e49ef18a9117caf (feat: support Claude Sonnet extended thinking models).