feat: Implement delivery forecasting

[bokelley]

Summary

AdCP beta.3 adds delivery forecasting - sellers provide budget curves showing how metrics scale with spend.

Types (adcp 3.3.0)

```python
from adcp.types import (
DeliveryForecast, ForecastMethod, ForecastPoint,
ForecastRange, ForecastRangeUnit, ForecastableMetric, DaypartTarget,
)
```

DeliveryForecast

• `points` - Forecasted delivery at budget levels (single point = standard, multiple = curve)
• `method` - How forecast was produced
• `currency` - ISO 4217 code
• `demographic` - Target demographic (Nielsen P18-49, BARB ABC1, etc.)
• `forecast_range_unit` - spend/reach_freq/weekly/daily/clicks/conversions
• `valid_until` - Forecast expiration

DaypartTarget

Time-of-day targeting:

• `days` - Days of week
• `start_hour` / `end_hour` - Time window (24hr format)
• `label` - e.g., "Morning Drive", "Prime Time"

Where Forecasting Fits

Forecasts could be returned:

• In `get_products` response (product-level forecast)
• As a separate `get_forecast` tool
• In `create_media_buy` response (what we expect to deliver)

Implementation

1. Determine where forecasts are surfaced in the API
2. GAM: Use GAM Forecast API for real forecasts
3. Mock: Generate simulated forecast curves
4. Daypart targeting: Add to TargetingOverlay handling

GAM Forecast API

GAM provides:

• `ForecastService.getAvailabilityForecast()` - Available impressions
• `ForecastService.getDeliveryForecast()` - Expected delivery
• Can forecast by line item or prospective line item

Acceptance Criteria

• Forecast data returned in appropriate responses
• GAM adapter calls Forecast API
• Mock adapter generates realistic curves
• DaypartTarget supported in targeting overlay
• Tests

[KonstantinMirin]

This feature requires DeliveryForecast, ForecastMethod, ForecastPoint, and related types available in adcp 3.3.0. Prerequisite: #1043 (adcp 3.3.0 upgrade).