Say why transpose is off, instead of naming the wrong cause

[thcp]

Closes #552

A client opening a networked StemDeck saw greyed-out key controls and, on the lane steppers, "Transpose needs Web Audio". Web Audio was working fine.

The cause, measured

AudioWorklet is a secure-context API. A browser on http://<lan-ip> is never given it, so the SoundTouch stage cannot be built at all. One page served three ways, in Chromium:

http://127.0.0.1:8777/     secureContext=true   audioWorklet=true
http://localhost:8777/     secureContext=true   audioWorklet=true
http://192.168.1.14:8777/  secureContext=false  audioWorklet=false

So the greying out is correct. This PR does not make transpose work over a network and does not claim to. It fixes what the app said about it, which was wrong in three separate ways.

What changed

The lane tooltip named the wrong thing. It blamed Web Audio, which was working, and sent people looking for a missing browser feature instead of at the URL they typed. It now says transpose is not available over a network connection.

The transport control said nothing at all. Three dead buttons still reading "Transpose up a semitone". It explains itself now, on the group rather than the buttons, since a disabled button does not reliably fire the pointer events a tooltip needs.

Speed was silently moving the key. This is the one nobody reported. Without the pitch stage the sources resample, so changing speed also changes the key, and the control gave no sign. It says so now.

Where the reason lives

pitchBlockedKey() is in pitchBus.js, for the same reason the input mapping is: the transport control and the lane steppers both have to answer this question, and two copies drift into two different answers. On an already-secure origin it still reports a plain worklet failure, which is then the only remaining explanation.

Deliberately not in the tooltip

There are real workarounds: an ssh tunnel to localhost, tailscale serve, an HTTPS reverse proxy. They are in the README troubleshooting section for anyone who goes looking. They are not in the tooltip, because answering a musician with port forwarding is a worse answer than admitting the limit.

The capability itself is tracked in #558, with the measurement showing createScriptProcessor is not secure-context gated and an honest account of what that path would cost.

i18n

Three keys added across all ten tables. ptPT overrides the new string, since European Portuguese says "ligação" rather than "conexão". mixer.key.unavailable is removed from all ten, dead once the lane tooltip started sharing the transport's answer.

Verification

playwright        90 passed
node tests/js     11/11
i18n audit        clean

Five e2e tests, including that a secure origin says something different and that a working transpose explains nothing at all.

Checked against the unpatched code: four of the five fail, and the one that passes is the one asserting the absence of a tooltip. So the tests catch the bug rather than merely describing the fix.