Fix chunk loss in the long streaming response with native response field

[TomeHirata]

Issue
There is an issue that some text content chunks are lost when there are more than 10 chunks for a string field, and the native response, such as Anthropic Citation, is used.

Root Cause
This was caused by the assumption that at most one streaming listener returns a chunk for the same chunk. This assumption is not true when native response, such as citation API, is used because a streaming listener buffers 10 chunks for string output fields, and sometimes the string chunk and native response chunk are produced simultaneously.

Fix
This PR fixes the issue by allowing multiple listeners to return a chunk for the same response chunk.

[Copilot]

Pull Request Overview

This PR fixes a bug where text chunks were lost in long streaming responses when using native response fields like Anthropic Citations. The issue occurred when a streaming listener's buffer contained more than 10 chunks, and both string chunks and native response chunks were produced simultaneously.

• Modified the streaming logic to handle multiple listeners returning chunks simultaneously
• Enhanced test coverage with more realistic scenarios involving 10+ chunks
• Fixed the assumption that only one listener can return a chunk at a time

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
tests/streaming/test_streaming.py	Enhanced test with 10+ chunks and added answer chunk verification
dspy/streaming/streaming_listener.py	Added guard to avoid empty tokens in buffer
dspy/streaming/streamify.py	Fixed logic to handle multiple simultaneous chunk returns

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}