Fix LangChain runtime issues and align JS integration with 1.x

[saschabuehrle]

Summary

• fix Python LangChain streaming by passing tags/metadata via runnable config in nested stream/astream calls
• make Python bind_tools() compatible with bound Runnable wrappers (e.g. RunnableBinding)
• keep low-risk tool-call policy explicit in Python tests (high threshold does not force verifier for low-risk tool calls)
• align @cascadeflow/langchain package deps with LangChain JS 1.x and add @langchain/langgraph for examples
• fix JS example/runtime issues:
  • remove invalid override modifiers under LangChain 1.x types
  • update LangGraph example to use Annotation.Root state definition
  • guard drafterQuality/latency formatting in langsmith-all-models example

Validation

• python -m pytest cascadeflow/integrations/langchain/tests -q
• pnpm --filter @cascadeflow/langchain build
• pnpm --filter @cascadeflow/langchain typecheck
• pnpm --filter @cascadeflow/langchain test
• live smoke with real API keys:
  • Python: examples/langchain_basic_usage.py, examples/langchain_streaming.py, examples/langchain_langgraph_multi_agent.py, examples/langchain_langsmith.py
  • TS: examples/basic-usage.ts, examples/streaming-cascade.ts, examples/lcel-pipeline.ts, examples/tool-risk-gating.ts, examples/model-discovery.ts, examples/langgraph-multi-agent.ts
  • cross-provider smoke (ChatAnthropic -> ChatOpenAI)