Add broadcast for stream open in process_stream_chunk

[TheRealNeil]

Fixes #289

Summary

Add broadcast_stream_open call in OpenAI::ChatProvider#process_stream_chunk to establish the streaming lifecycle

Root cause

The OpenAI ChatProvider#process_stream_chunk never called broadcast_stream_open, so the streaming flag was never set to true. When streaming finished, broadcast_stream_close checked return unless streaming and returned early — meaning on_stream_close callbacks never fired. This affected both OpenAI and Ollama (which inherits from the OpenAI provider).

The Anthropic and mock providers already called broadcast_stream_open correctly; the OpenAI provider was the odd one out.

Fix

Add broadcast_stream_open at the top of process_stream_chunk, mirroring the pattern in the Anthropic and mock providers. This ensures the full stream lifecycle (open → update → close) fires correctly.

Test plan

• Verify Ollama streaming now triggers on_stream_close callbacks
• Verify OpenAI streaming open/close lifecycle works end-to-end
• Confirm Anthropic and mock provider streaming is unaffected
• Confirm broadcast_stream_open is idempotent when called on every chunk

[Copilot]

Pull request overview

Adds the missing stream “open” broadcast to the OpenAI chat streaming path so the stream lifecycle (open → update → close) is correctly established and on_stream_close callbacks can fire (fixing behavior for both OpenAI and Ollama, which inherits from it).

Changes:

• Call broadcast_stream_open at the start of OpenAI::ChatProvider#process_stream_chunk.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Add a regression test that asserts the OpenAI (and ideally Ollama) provider streaming lifecycle emits :open and :close events when stream: true. Right now the change fixes a lifecycle bug, but there isn't provider-level test coverage to prevent broadcast_stream_open from being accidentally removed again (similar to the existing mock provider streaming test).