Module streaming 

In-flight streaming assistant preview.

Renders partial text via full MessageFormatter. Uses a thread-local parse cache to avoid re-running the markdown formatter on every token: while the streaming text grows by fewer than [STREAM_REPARSE_THRESHOLD] bytes since the last parse, the cached lines are reused as-is.

Functions

push_streaming_preview

Append a streaming preview block when the app is actively receiving text.

reset_stream_parse_cache

Reset the streaming parse cache. Call when a new assistant turn begins (e.g. streaming_text was cleared) so a shrunken buffer doesn’t keep reusing stale parsed lines.