Build WearOS and watchOS companion apps for sleep stage detection

## Summary
Build native companion apps for WearOS and watchOS that run during sleep sessions to provide real-time accelerometer data and optimized vitals streaming for improved sleep stage classification.

## Background
Currently we rely on Health Connect/HealthKit for sleep data, which has limitations:
- **HRV data** - Fitbit doesn't sync HRV to Health Connect
- **Movement detection** - No accelerometer access through Health Connect
- **Data latency** - Health data is synced periodically, not real-time
- **Time alignment** - HR samples often don't overlap with sleep stage windows

## Proposed Solution

### WearOS Companion App
- **Platform**: Kotlin, Jetpack Compose for Wear OS
- **Sensors**: Accelerometer (10Hz), Heart Rate (continuous), HRV if available
- **Communication**: DataClient/MessageClient API to phone app
- **Battery**: Foreground service with wake lock, optimized sampling

### watchOS Companion App
- **Platform**: Swift, SwiftUI
- **Sensors**: Accelerometer, Heart Rate via HealthKit workout session
- **Communication**: WatchConnectivity framework
- **Battery**: Extended runtime session for sleep tracking

### Data Protocol
```typescript
interface WatchData {
timestamp: number;
heartRate?: number;
hrvMs?: number;
accelerometer?: { x: number; y: number; z: number };
movementIntensity?: number; // computed on-watch to reduce bandwidth
}
```

### Features
1. **Real-time streaming** during active sleep sessions
2. **On-watch movement computation** to reduce data transfer
3. **Local buffering** if phone connection lost
4. **Battery-efficient sampling** (reduce accelerometer rate during low-movement periods)
5. **Sleep stage inference on watch** as fallback

## Implementation Phases

### Phase 1: WearOS MVP
- [ ] Basic foreground service with accelerometer sampling
- [ ] Heart rate continuous monitoring
- [ ] DataClient communication to phone
- [ ] Phone-side receiver and integration with sleep.ts

### Phase 2: watchOS MVP
- [ ] HealthKit workout session for extended runtime
- [ ] WatchConnectivity message passing
- [ ] Phone-side receiver for iOS

### Phase 3: Optimization
- [ ] On-device movement intensity computation
- [ ] Adaptive sampling rates
- [ ] Battery usage optimization
- [ ] Offline buffering and sync

## Acceptance Criteria
- [ ] Watch app runs reliably for 8+ hour sleep session
- [ ] Movement data available within 1 second of occurrence
- [ ] Battery drain < 15% overnight on watch
- [ ] Improved sleep stage classification accuracy vs Health Connect alone

## Technical Notes
- WearOS: Use `SensorManager` for accelerometer, `HealthServicesClient` for HR
- watchOS: Use `CMMotionManager` for accelerometer, `HKWorkoutSession` for HR
- Consider using Protocol Buffers for efficient data serialization

## Related
- Improves sleep stage classification accuracy
- Enables movement-based awake detection
- Provides HRV data that Fitbit doesn't sync to Health Connect