Fix: Bedrock Application Inference Profile (AIP) is not streaming response with ChatBedrockConverse

[0x-fang]

Description

This PR fixed the streaming issue with ChatBedrockConverse when input is Application Inference Profile (AIP). The issue happened because langchain-aws fails to identify the foundation model used in AIP (e.g., arn:aws:bedrock:us-east-1:111111484058:application-inference-profile/c3myu2h6fllr), therefore, it cannot set the streaming_support flag for the AIP correctly. So the entire response was returned to user as a whole rather than streaming the response back.

Solution

We have to create a Bedrock client to call Bedrock get_inference_profile control plane API, and parse the foundation model id from the response. However, the existing set_disable_streaming function works on the raw user input before the ChatBedrockConverse object is instantiated, so we have to extract the logic of determining streaming_support for models into a common function _get_streaming_support and invoke it in both places (i.e., the original place that works on raw user input, and the new place that works on the resolved model Id from get_inference_profile API call).

Issue

#538

Test

• add new unit tests in libs/aws/tests/unit_tests/chat_models/test_bedrock_converse.py and all passed
• run through the integration test for ChatBedrockConverse and all passed except one existing failure
  • The main branch has failed this integration test already, which is not caused by new change.

  FAILED tests/integration_tests/chat_models/test_bedrock_converse.py::test_structured_output_tool_choice_not_supported - assert 1 == 0
  

[michaelnchin]

Integ CI passes: https://github.com/langchain-ai/langchain-aws/actions/runs/16761621041

[michaelnchin]

Thanks @0x-fang !