LLM streaming events are broadcast to all Electron windows instead of only the consuming window

[Ayushpal2006]

Description

During text and screenshot-based LLM streaming, every response start and chunk event is broadcast to all registered Electron windows using broadcastToAllWindows().

The relevant events are:

• transcription-llm-response-start
• transcription-llm-response-chunk
• transcription-llm-response

However, the streaming handlers appear to be implemented only in the chat renderer. This means unrelated windows, including hidden windows such as settings and the main overlay, receive every streamed chunk.

Because an LLM response can contain many chunks, this creates unnecessary IPC messages and per-event logging.

Relevant code

In main.js, both image and text processing use:

windowManager.broadcastToAllWindows(
  "transcription-llm-response-chunk",
  {
    messageId,
    delta
  }
);