fix: handle forecast shape correctly for quantile calculation

- Add logic to detect forecast shape (samples, time) vs (time, samples)
- Compute quantiles along correct axis based on shape
- Add median shape validation logging
- This should fix the dimension mismatch error
- Bump to v1.0.7

src/forecasting/chronos_inference.py

@@ -2,7 +2,7 @@
 """
 Chronos-2 Inference Pipeline
 Standalone inference script for HuggingFace Space deployment.
-FORCE REBUILD: v1.0.6
+FORCE REBUILD: v1.0.7
 """
 
 import os
@@ -188,11 +188,26 @@ class ChronosInferencePipeline:
                 forecast_numpy = forecast.numpy()
                 print(f"[DEBUG] Forecast shape: {forecast_numpy.shape}, Expected: ({num_samples}, {prediction_hours})", flush=True)
 
+                # Compute quantiles along the sample axis (axis=0 for shape (num_samples, time))
+                # If shape is transposed (time, num_samples), use axis=1
+                if forecast_numpy.shape[0] == prediction_hours:
+                    # Shape is (time, samples) - use axis=1
+                    median = np.median(forecast_numpy, axis=1)
+                    q10 = np.quantile(forecast_numpy, 0.1, axis=1)
+                    q90 = np.quantile(forecast_numpy, 0.9, axis=1)
+                else:
+                    # Shape is (samples, time) - use axis=0
+                    median = np.median(forecast_numpy, axis=0)
+                    q10 = np.quantile(forecast_numpy, 0.1, axis=0)
+                    q90 = np.quantile(forecast_numpy, 0.9, axis=0)
+
+                print(f"[DEBUG] Median shape: {median.shape}, Expected: ({prediction_hours},)", flush=True)
+
                 # Store results
                 results['borders'][border] = {
-                    'median': np.median(forecast_numpy, axis=0).tolist(),
-                    'q10': np.quantile(forecast_numpy, 0.1, axis=0).tolist(),
-                    'q90': np.quantile(forecast_numpy, 0.9, axis=0).tolist(),
+                    'median': median.tolist(),
+                    'q10': q10.tolist(),
+                    'q90': q90.tolist(),
                     'inference_time_s': time.time() - border_start
                 }