feat: Add tool support to AnthropicMultiModal

[rushilsheth]

Description

This PR adds support for handling tool responses in the AnthropicMultiModal class. Currently, when using tools with Claude 3, the LlamaIndex wrapper expects a text response but fails when receiving a tool response instead. This change enables handling both response types seamlessly.

Key changes:

• Modify _complete method to handle both tool and text responses
• Add unit tests to verify tool response handling
• Update documentation with tool usage examples and best practices

Type of Change

• New feature (non-breaking change which adds functionality)

How Has This Been Tested?

• I added new unit tests to cover this change

Added tests for:

• Tool response handling
• Standard text response compatibility
• Error cases
• Response format validation

Detailed Changes

Code Changes

Modified _complete method to detect and handle tool responses:

def _complete(self, prompt: str, image_documents: Sequence[ImageNode], **kwargs: Any) -> CompletionResponse:
    # ... existing setup code ...
    
    content = response.content[0]
    if hasattr(content, "input"):
        # Handle tool response
        text = str(content.input)
    else:
        # Handle standard text response
        text = content.text
        
    return CompletionResponse(
        text=text,
        raw=response,
        additional_kwargs=self._get_response_token_counts(response),
    )

New Tests

def test_anthropic_tool_response():
    """Test tool response handling"""
    # ... test implementation ...

Checklist:

• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I ran make format; make lint to appease the lint gods

New Package?

• No

[logan-markewich]

@rushilsheth I'm happy to merge this, but tbh I recently updated the normal Anthropic class to work with images, albiet with our newer chat-message format (we are slowly rolling out support for this, so far openai, azure openai, and anthropic support this)

from llama_index.core.llms import (
    ChatMessage,
    ImageBlock,
    TextBlock,
    MessageRole,
)

msg = ChatMessage(
    role=MessageRole.USER,
    blocks=[
        TextBlock(text="Describe the images as an alternative text"),
        ImageBlock(path="some_file.png", image_mimetype="image/png"),
    ],
)

response = llm.chat(messages=[msg])

This means you can use the same llm class for multimodal, tool calling, etc, (assuming you use the lower-level llm operations I guess)

[rushilsheth]

It would be great if you could merge this. We'll be sure to migrate our internal code base to the new base class after some testing

[logan-markewich]

ngl this PR is pretty hacky. But since this class

• a isn't handling tool calls properly anyways
• isn't meant to be maintained much further

Should be ok