4.5.2

Fixed

• AI: /rewrite and /edit no longer fail with reasoning models (GPT-5 family).
  Reasoning models spend their hidden "thinking" tokens inside the response token limit.
  On long notes, the chat-sized limit (default 1500) was entirely consumed by reasoning
  before the first visible character, so the final "Rewrite note" step came back empty
  and Znote reported "No rewrite produced" — while non-reasoning models like gpt-4.1
  worked fine. Rewrite generations now use a dedicated 16k-token budget, sized for
  reasoning plus a complete note body. This also removes the risk of a long rewrite
  being silently truncated by the chat response-length setting.
  Models with a smaller context window are unaffected: if they can't accept the raised
  budget, the cap is automatically dropped and the request retried.