feat: audio upload

[onkoe]

if a user uploads an audio file, the script can use that instead of requiring video files.

my current process is kinda funny:

1. record giant 40 GiB file
2. extract audio and compress with Audacity (Tenacity)
3. add a black screen to the video with ffmpeg: ffmpeg -f lavfi -i color=c=black:s=1280x720:r=5 -i input.mp3 -crf 0 -c:a copy -shortest output.mp4
4. upload that to the google colab
5. wait for it to extract the audio
6. then, finally, it can start the transcription process

that's not great... but yes, it does technically work