v0.3.1

Tested on WSL2 with Ubuntu 24.04. The regression session comes out identical to the reference, the deck's self-test passes, and setup was run from scratch into a new folder.

• The app's sound is paced by the deck's sound card, not the host's clock. On WSL the clock ran about 7 % fast, so time in the app ran fast too, and the deck cut 400 ms of sound every few seconds to catch up: a skipped beat about once a pattern, and sound behind the screen. --no-pace brings back the old pacing, and DJS_AUDIO_STATS=1 prints the speeds every 10 s.
• The deck keeps 250 ms of sound waiting by default (--audio-buffer, 150 before), so a short stall of the app no longer runs it dry.
• The sound's cuts, and the times the deck asked for sound late, are the status line's hint for the sound device, kept up to date while it is hovered. The device's name turns orange once either has happened.
• The docs keep to what runs the emulator. The hardware and power pages are gone, and the update file page is shorter.
• The regression check times its sequence's STOP by the sound the app has played, not by this machine's clock. WSL's clock runs a few percent away from Windows', by a different amount from week to week, and STOP fell into the last sound's tail, so the check failed now and then with nothing wrong. Session.frames_heard() counts that sound for other scripts.