VScrybe 0.3.0

First public release, now also for Windows and macOS.

Added

• Settings window (gear beside About) with Models, Hugging Face, This computer, Appearance and About tabs. Add
  Whisper models from a Hugging Face list or a folder on this computer; a downloading model shows its progress.
• Windows and macOS builds, for both Apple silicon and Intel Macs, and an AppImage for any Linux distribution.

Changed

• vScribe is now VScrybe. Notes, settings and downloaded models move to the new name's folders the first
  time it starts.
• Audio is decoded by FFmpeg built into VScrybe, so nothing has to be installed on any system. It is also more
  accurate than the GStreamer pipeline it replaces (stereo was mixed down about 3 dB too quiet).
• Only one model is kept in memory at a time, and uploads are written to disk instead of held in memory.

Fixed

• On Windows, recording showed a second, browser-style microphone prompt naming the app's local address and
  port. VScrybe's own consent dialog is now the only one.
• Long stretches of speech could be left out when Whisper stopped early in a 30-second window. Transcription now
  resumes where it stopped, retries unreliable output and discards repetition loops.
• Files with a few undecodable packets (some .amr recordings) failed entirely; those packets are now skipped.
• Download progress stayed at 0% until a model finished, and a second download waited behind the first.
• Gated Hugging Face models failed with "check your internet connection"; they are now marked and refused.
• Dialogs covered the window controls, and clicking a "Transcription finished" notification didn't open the app.
• Quitting while a note was being transcribed could crash the app. It now stops cleanly, and the note picks up
  again the next time VScrybe starts.

Full Changelog: https://github.com/Invalid8/vscrybe/commits/v0.3.0