Add LangGraph auto-instrumentation for Braintrust hook imports

[Stephen Belanger (Qard)]

Summary

• Add LangChain/LangGraph auto-instrumentation to braintrust/hook.mjs so callback handlers are injected automatically without requiring @braintrust/langchain-js imports.
• Keep the manual @braintrust/langchain-js API working while reusing the same span semantics through a vendored callback-handler surface.
• Add config support for langchain, langchain-js, and langgraph disable aliases and export the new LangChain instrumentation configs.
• Add a real provider-backed e2e scenario that verifies --import braintrust/hook.mjs captures LangGraph task spans and nested OpenAI spans end to end.

Testing

• Passed focused Vitest coverage for the new LangChain plugin, config mapping, registry integration, and existing @braintrust/langchain-js tests.
• Passed the new provider-backed LangGraph e2e scenario repeatedly with OPENAI_API_KEY set.
• Passed the repo build, and the touched files were formatted successfully.

Fixes #1889

[Luca Forstner (lforst)]

Let's also add e2e test ci summary for this.