fix(chat): expose stream_batch content for every response

[latent-9]

Description

stream_batch(n) wraps all n options in Response objects that share one underlying proto, but only responses[0].process_chunk(chunk) is called for every incoming chunk. Chunk content is buffered per output index inside the first response, and only materialized into the shared proto when responses[0].content is read, or when the telemetry span closes.

A caller that reads responses[i].content for i > 0 during streaming, or before responses[0], gets an empty string for the whole stream. This is the natural way to stream per-option content:

async for responses, _ in chat.stream_batch(2):
    print(responses[1].content)

With this code responses[1].content is "" on every iteration, even though the server streams the content. The fix materializes the buffered content into the shared proto after every chunk, so each response reflects the streamed content regardless of read order.

Changes

• src/xai_sdk/aio/chat.py: sync content buffers to the shared proto after each chunk in stream_batch
• src/xai_sdk/sync/chat.py: same fix for the sync client
• Regression tests in tests/aio/chat_test.py and tests/sync/chat_test.py that read responses[1].content during the stream. They fail on the current code and pass with this change.
• Changelog entry under Unreleased.

Type of Change

• Bug fix

Related Issue

N/A