WhisperASR v0.6.1

• Fixed a crash risk when switching models while a transcription or live session was in progress — model loading now serializes with inference
• Removing a transcription no longer deletes imported audio files; app-made recordings go to the Trash instead of being permanently deleted
• The live transcript is kept as a transcript-only item if the recording file fails to save, instead of being lost
• Renaming no longer breaks an item when the file move fails (e.g. the name already exists)
• Tapping a transcript segment right after opening an item now seeks correctly instead of playing from the start
• Faster sidebar search on large libraries, and lower CPU/battery use while recording and during playback
• Transcript search handles unusual Unicode text safely and highlights matches consistently
• The API server rejects uploads over 1 GB with a proper 413 error instead of buffering them in memory
• Drag-and-drop now accepts MOV, WebM, MKV and Opus files, matching the file picker
• The recording clock and audio-stall watchdog keep running while the "meeting ended" alert is open
• Clearer error message when no speech recognition model is installed; fixed a memory leak in model downloads