chore: phone orchestra example

apps/common-app/orchestra-server/.gitignore

@@ -0,0 +1,4 @@
+samples/*
+!samples/.gitkeep
+midi/*
+!midi/.gitkeep

apps/common-app/orchestra-server/README.md

@@ -0,0 +1,97 @@
+# Phone orchestra server
+
+Companion server for the **Orchestra** demo (`src/demos/Orchestra/`). Phones
+connect over WebSocket, estimate the server clock (NTP-style ping/pong) and
+schedule their part of the score with `start(when)`, so a room full of phones
+plays one piece together.
+
+```sh
+node server.ts            # Node 22.18+ (type stripping), no dependencies
+PORT=9000 node server.ts  # custom port (default 8765)
+```
+
+The server prints its LAN addresses: open the `http://…` one in a browser for
+the dashboard (start/stop, tempo, score, per-phone part assignment,
+sync quality). Phones join by pointing their camera at the dashboard's QR
+code: it encodes `audioapi-example://orchestra?url=ws://<host>:8765`, which
+opens the app on the Orchestra screen and connects. The `ws://…` address can
+also be typed in by hand.
+
+The QR code is rendered by `qrcodegen.ts`, vendored from
+[nayuki/QR-Code-generator](https://github.com/nayuki/QR-Code-generator) (MIT)
+and adapted to the erasable-syntax subset Node's type stripping accepts; its
+output was checked identical to the upstream build for inputs spanning
+versions 1–8.
+
+Phones and the laptop must share a network. Conference Wi-Fi is the usual
+failure point — bring a router, and watch the RTT column on the dashboard.
+
+Scores live in `score.ts`; the wire format in `../src/demos/Orchestra/protocol.ts`.
+
+## Recorded instruments
+
+Parts with an `instrument` play recorded notes instead of the synth: every
+General MIDI instrument of the FluidR3_GM soundfont, one mp3 per semitone
+(~22 KB), or `drums`, an acoustic kit (kick, snare, clap, closed and open
+hi-hat, two toms, crash, ride; a few takes each, picked at random) for parts
+whose notes are General MIDI percussion keys. The kit's key mapping lives in
+`drumKit.ts`. At startup the server downloads every note the scores use into
+`samples/` (kept out of git) and serves it at `/samples/<instrument>/<note>.mp3`,
+so phones need only the local network; notes nobody asked for yet are
+downloaded on first request. Run the server once with internet access before
+a show. Without it, phones fall back to the synth voices.
+
+Phones load the notes of their own part when it is dealt, and play the synth
+for any note that has not loaded yet. The phone's Voice row can switch any
+part to piano, strings, flute, trumpet, harp or marimba, or back to the synth.
+
+Sample credits: FluidR3_GM by Frank Wen (MIT), rendered to per-note mp3 by
+Benjamin Gleitzman in [midi-js-soundfonts](https://github.com/gleitz/midi-js-soundfonts)
+(CC BY 3.0). Drums from the [Versilian Community Sample Library](https://github.com/sgossner/VCSL)
+by Versilian Studios (CC0).
+
+## Importing MIDI files
+
+Drop Standard MIDI Files (`.mid`, format 0 or 1) into `midi/` and restart the
+server (or run `node --watch server.ts`). Each file becomes a score named after
+the file. Only import music you have the right to use.
+
+At startup the server prints every imported part, for example:
+
+```
+MIDI My Song.mid → "My Song" (midi-my-song), 120 bpm, 128 beats
+  part "Lead": 212 notes, midi 64-84, pluck
+  part "Bass": 160 notes, midi 33-45, transposed +24, pluck
+  part "Drums (ch 10)": 410 notes, midi 36-51, kit
+```
+
+- Each track (or each channel of a multi-channel track) becomes one part.
+- Each part plays the recorded instrument its General MIDI program selects
+  (piano when the file sets none).
+- Channel 10 becomes a `kit` part on the recorded drum kit; keys the kit has
+  no piece for (cowbell, congas, …) are dropped.
+- Parts lower than midi 55 move up whole octaves so phones can play them.
+- Only the first tempo is used; tempo changes are reported as a warning.
+
+To choose voices and how many phones play each part, add `My Song.json` next to
+`My Song.mid`, using the part names the server printed:
+
+```json
+{
+  "title": "My Song",
+  "bpm": 118,
+  "parts": {
+    "Lead": { "instrument": "violin", "maxPlayers": 1 },
+    "Bass": { "timbre": "pluck", "transpose": 12 },
+    "Strings": { "skip": true }
+  }
+}
+```
+
+`instrument` takes any name from
+[the FluidR3_GM list](https://gleitz.github.io/midi-js-soundfonts/FluidR3_GM/names.json)
+and replaces the one the file selects. Timbres (the synth fallback): `bell`,
+`pluck`, `pad`, `lead`, `hat`, `clap`, `kit`, `click`.
+`maxPlayers` caps how many phones are dealt the part automatically (the rest
+fill the uncapped parts). `transpose` (semitones) replaces the automatic octave
+shift. Files that fail to parse are reported and skipped.