feat: Support Anthropic Claude prompt caching key "cache_control"

[serialx]

Anthropic models require explicit cache control blocks for the caching to work. And the way they achieve this is by adding an additional key to the text block like:

    "system": [
      {
        "type": "text",
        "text": "You are an AI assistant tasked with analyzing literary works. Your goal is to provide insightful commentary on themes, characters, and writing style.\n"
      },
      {
        "type": "text",
        "text": "<the entire contents of Pride and Prejudice>",
        "cache_control": {"type": "ephemeral"}
      }
    ],
    "messages": [
      {
        "role": "user",
        "content": "Analyze the major themes in Pride and Prejudice."
      }
    ]

ref: https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching

Since "cache_control" key is not in official OpenAI API specs, there are various type errors I needed to maneuver through. Code is not ideal, but I tried to create a patch with minimal impact.

With this patch, we can inject cache control keys to the input like:

input_items: list[TResponseInputItem] = []
input_items.append({"role": "user", "content": "Hello, world", "cache_control": {"type": "ephemeral"}})
result = await Runner.run(current_agent, input_items)

This solves #905

[github-actions]

This PR is stale because it has been open for 10 days with no activity.

[serialx]

I've solved this with a custom model that inherits LitellmModel. Closing because this is a very inelegant approach.

[serialx]

Here's my code if anyone is interested: https://gist.github.com/serialx/31dd60db626b3dcb646dcd8369fd2552

[ZXTFINAL]

Here's my code if anyone is interested: https://gist.github.com/serialx/31dd60db626b3dcb646dcd8369fd2552

Thanks! It helps a lot.