v1.4.0 - OpenAI-compatible providers & clipboard preservation

What's new

The biggest release since 1.0 - with two major community contributions. 🎉

Features

• 🤖 OpenAI-compatible local providers - the voice assistant now works with
  llama.cpp, LM Studio and any
  OpenAI-compatible local server, not just Ollama. Pick the provider in Settings;
  models are discovered automatically, and each provider keeps its own URL and
  model. Implements #16. (thanks @aladin7!)
• ⌨️ Bare modifier hotkeys - you can now bind a lone modifier such as Right
  Ctrl as your dictation or assistant hotkey via config/settings. Fixes #19
  (thanks @MaxiKingXXL for the precise report).

Fixes

• 🖼️ Dictation no longer destroys your clipboard - screenshots, copied files
  and other non-text formats are preserved and restored after the paste. If you
  copy something new while WritHer is pasting, your fresh copy wins. (thanks @aladin7!)
• 👻 Quit really quits, for real this time - v1.3.2 could still leave an
  invisible process alive after quitting from the tray (making the next launch
  say "already running"). The shutdown now runs on the right thread and always
  terminates. If you hit this on v1.3.2: Task Manager → end WritHer.exe, then upgrade.
• 🚀 Startup status now reads the real model cache, so "Downloading speech
  model…" only appears when a download is actually happening.

Full Changelog: v1.3.2...v1.4.0