[lana-20]

Fixes #334.

audio.sh passed 1 of 6 on an Android 15 emulator started by hand: one steady tone was heard as two sounds.

Cause

The issue guessed a dropout. That was wrong: the tone sounds unbroken, and packets were neither late nor short. The waveform has pairs of breaks 33 samples apart, with the tone's phase jumping across them. An FFT across them read 430 Hz for 440, past the 2% that holds one sound together.

The breaks are the emulator's host audio output. The same app and tones gave 151 breaks in 48 s when started by hand, and none from mobium boot, which passes -no-audio. They cluster after audio starts, which is why a loop of captures rarely split and the check usually did.

What changes

• steadyPitch cuts a window at each break and takes the median of the stretches. Stretches that are not one sound are two notes, read whole as before, so a melody and a real pitch change still split.
• docs/SETUP.md boots with mobium boot, and says what a hand start costs.
• -port leaves an emulator with no control port, measured on emulator 37.1.11. SETUP.md said to pass it, and the refusal blamed -no-grpc alone.
• audio.sh took a tone's first instant for a sound in the pause, failing both versions now and then. A sound inside the pause, and a too-short pause, still fail it.

Measured

audio.sh by hand: main 1 of 6, this 6 of 6. On -no-audio, 4 of 4, and Android 17 passes. audio-sim.sh passes.

Real captures, started by hand:

Level	Captures	Split on main	Split here
-42 dBFS	36	1	0
-46 dBFS	16	2	0
-49 dBFS	16	3	0
-62 dBFS	14	5	2

Not settled here

• At -62 dBFS, the quietest step that sounds, a tone started by hand still splits 2 in 14: breaks that small go unseen.
• No tone from anything but MobiumApp was captured: Chrome played nothing. The evidence against the app's module is the boot comparison.
• Measured on one Mac only, and not on a phone, which captures no samples.