v3.3.0

- Add a second reply lane. A fast, no-thinking model drafts an instant reply
  suggestion in about a second while the considered Reply model keeps thinking,
  so a first draft appears without waiting for the slower answer. Both lanes run
  on the same finished turn and share supersede, cancellation, and the
  local-speech suppression that blocks a reply while you are still talking.
- Default the fast lane to groq-gpt-oss-120b with reasoning off. Make the
  considered Reply lane think, defaulting it to cerebras-glm-4.7 at medium
  reasoning instead of no reasoning.
- Add a Fast reply model picker and its own live feed panel to the web console.
  Migrate saved browser settings from schema 2 to schema 3 by back-filling the
  new lane with its default model and reasoning.
- Add a GitHub sponsors funding link.
- Move CI checks to the shared code-workflow and grant it security-events
  write access.