Add code execution middleware

[pokey]

• Migrate to be a PR against langchain-ai once feat(langchain/createAgent): Add support for modelSettings to middleware langchain-ai/langchainjs#9110 is merged
• Add a proper description
• Add todo to switch to using BaseStore instead of having new FileSystemProvider interface, and default to using graph if no store provided
• Probably export codeExecutionAnthropic instead of codeExecution. Internally can just wrap existing impl. Something like:

  export const codeExecutionMiddlewareAnthropic = (fileProvider) => codeExecutionMiddleware(
    new AnthropicContainerProvider(),
    fileProvider
  );
  
  // ...then clients do:
  
  const middleware = codeExecutionMiddlewareAnthropic(
    new MemoryFileProvider()
  );

  This gives us flexibility to switch to anthropic-specific and openai-specific middleware impl in the future because current container abstraction is awkward
• Note about forking off file system middleware, potentially using one from deepagent
• Then current codeExecution middleware abstraction goes away and anthropic-specific and openai-specific middleware just use file system middleware. Rest of stuff can become utility functions?
• Daytona becomes example
• Look into intermittent issue with extra tools example. Something about constructing paths is confusing the llm ie whehter or not to include the /tmp/ prefix. See https://smith.langchain.com/public/79d05f34-2924-4707-b0ae-73b8da3d9a9d/r

[pokey]

Prob want a way for middleware to shut itself off if the model is unsupported, eg Anthropic middleware with OpenAI model

[pokey]

Closing in favour of langchain-ai#9209