Add AWS Bedrock model provider support

[ystarikovich]

Summary

langmem currently ships with langchain-openai and langchain-anthropic as dependencies, but has no built-in support or documentation for using AWS Bedrock models via langchain-aws.

Many production workloads run on AWS and use Bedrock as their LLM gateway. Since langmem's internal functions use init_chat_model() from langchain, Bedrock models should work when passing a string like "bedrock:anthropic.claude-3-5-sonnet-20241022-v2:0" — but this is neither documented nor tested.

Request

• Add langchain-aws as an optional dependency:

  [project.optional-dependencies]
  aws = ["langchain-aws>=0.2.0"]

• Add documentation / examples showing how to use Bedrock models with langmem
• Validate that core features (memory extraction, prompt optimization, etc.) work with ChatBedrock / ChatBedrockConverse

Example usage

from langchain_aws import ChatBedrockConverse
from langmem import create_memory_store_manager

model = ChatBedrockConverse(model_id="anthropic.claude-3-5-sonnet-20241022-v2:0")
manager = create_memory_store_manager(model)

or via string identifier:

manager = create_memory_store_manager("bedrock:anthropic.claude-3-5-sonnet-20241022-v2:0")