fix: reduce context window to 1125 hours (1.5 months) for A100-80GB

- Previous 1,800h context caused OOM on A100-80GB (needed 93 GiB > 79 GiB available)
- 512h worked on 28GB GPU, so 1125h should fit on 80GB with headroom
- Estimated memory: ~48 GiB (31 GiB headroom)
- Updated both dynamic_forecast.py and chronos_inference.py

Co-Authored-By: Claude <[email protected]>

src/forecasting/chronos_inference.py

@@ -132,7 +132,7 @@ class ChronosInferencePipeline:
         run_date: str,
         borders: Optional[List[str]] = None,
         forecast_days: int = 7,
-        context_hours: int = 2160,  # 2,160 hours = 90 days (3 months, optimized for 96GB VRAM)
+        context_hours: int = 1125,  # 1,125 hours = 46.9 days (1.5 months, fits A100-80GB)
         num_samples: int = 20
     ) -> Dict:
         """

src/forecasting/dynamic_forecast.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python3
 """
-Dynamic Forecast Module v1.7.0 - Extended Context Window (75 Days / 2.5 Months)
+Dynamic Forecast Module v1.8.0 - Context Window (47 Days / 1.5 Months)
 Time-aware data extraction for forecasting with run-date awareness.
 
 Purpose: Prevent data leakage by extracting data AS IT WAS KNOWN at run time.
@@ -8,7 +8,7 @@ Purpose: Prevent data leakage by extracting data AS IT WAS KNOWN at run time.
 Key Concepts:
 - run_date: When the forecast is made (e.g., "2025-09-30 23:00")
 - forecast_horizon: Always 14 days (D+1 to D+14, fixed at 336 hours)
-- context_window: Historical data before run_date (1,800 hours = 75 days / 2.5 months, optimized for A100-80GB)
+- context_window: Historical data before run_date (1,125 hours = 47 days / 1.5 months, fits A100-80GB)
 - future_covariates: ALL 2,514 features (leveraging Chronos-2 past-only masking)
   * 603 full-horizon (known future)
   * 12 partial D+1 (masked D+2-D+14)
@@ -39,7 +39,7 @@ class DynamicForecast:
     def __init__(
         self,
         dataset: pl.DataFrame,
-        context_hours: int = 1800,  # 1,800 hours = 75 days (2.5 months, optimized for A100-80GB)
+        context_hours: int = 1125,  # 1,125 hours = 46.9 days (1.5 months, fits A100-80GB)
         forecast_hours: int = 336  # Fixed at 14 days
     ):
         """