feat: connect synth runtime to canonical voices

[ChuxiJ]

Summary

• switch canonical subtractive playback from basic Tone.Synth voices to Tone.MonoSynth so filter, filter envelope, detune, and fat-oscillator unison settings affect real playback
• switch canonical FM playback from legacy fallback presets to real Tone.FMSynth voices with mapped carrier/modulator settings
• add runtime-spec regression tests so the engine-side parameter mapping stays stable as the editor evolves
• update the FM editor note to reflect that live playback now uses the FM runtime path

Verification

• npx tsc --noEmit
• npx vitest run src/engine/__tests__/SynthEngine.test.ts tests/unit/SynthInstrumentEditor.test.tsx tests/unit/PianoRoll.test.tsx tests/unit/VirtualKeyboard.test.tsx tests/unit/trackPresets.test.tsx src/utils/__tests__/trackInstrument.test.ts tests/unit/pianoRollContextMenu.test.tsx tests/unit/sessionLaunchModes.test.ts tests/unit/sessionLegato.test.ts
• npm run build

Notes

• This is a stacked PR on top of feat: add piano roll synth editor shell #1007.
• LFO and some higher-level character controls are still stored canonically but not fully rendered in the runtime yet; this slice focuses on the highest-signal playback gaps first.

[ChuxiJ]

Triage: Base of synth chain — This is the base PR for the synth improvement stack (#1015 → #1018 → #1019 → #1060). Needs rebase and review before the rest of the chain can progress.

---

Generated by Claude Code