Support thinking config for Bedrock hosted Claude models

[varunpandey23]

Changes:

• Upgrade langchain-aws and use ChatBedrockConverse to support extended and adaptive thinking
• Extract text content from model response, as Claude models with thinking returns a list as the content.

Example model configuration:

{
    "llm_provider": "aws_bedrock",
    "llm_model": "us.anthropic.claude-sonnet-4-6",
    "llm_api_key": "not_needed",
    "temperature": 1.0,
    "max_tokens": 16000,
    "reasoning": {"type": "adaptive"}
}