fix(anthropic): avoid Pydantic warnings for message_stop events

[opieter-aws]

Motivation

When using AnthropicModel, Pydantic emits PydanticSerializationUnexpectedValue warnings when the Anthropic SDK returns ParsedTextBlock objects in the message content during message_stop events. This is because ParsedTextBlock is a Pydantic BaseModel that cannot be cleanly serialized against the TypedDict union variants in the ContentBlock type.

The warnings do not break functionality but create noise in user applications and may mask other legitimate warnings.

Resolves #1746

Public API Changes

No public API changes.

Technical Approach

Instead of suppressing warnings with warnings=False, this fix addresses the root cause by avoiding unnecessary serialization entirely. For message_stop events, we build the dict directly from event.message.stop_reason instead of calling event.model_dump(), since format_chunk only needs the stop_reason field for this event type.

# Before (causes warnings - serializes entire message including content)
async for event in stream:
    if event.type in AnthropicModel.EVENT_TYPES:
        yield self.format_chunk(event.model_dump())

# After (fixes root cause - only extracts what we need)
async for event in stream:
    if event.type in AnthropicModel.EVENT_TYPES:
        if event.type == "message_stop":
            yield self.format_chunk({
                "type": "message_stop",
                "message": {"stop_reason": event.message.stop_reason},
            })
        else:
            yield self.format_chunk(event.model_dump())

This approach is cleaner than warning suppression because it:

• Fixes the actual problem rather than hiding it
• Avoids masking other legitimate Pydantic warnings
• Is more efficient (less data serialized)
• Makes the code explicit about what data is needed

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!