feat: support %_generation column in solar profile CSV

- Add column %_generation (percentage, 0-100%) to solar profile CSV
- Update loader to read %_generation and convert to normalized fraction
- Fallback to old E_Grid_Kwp format for backwards compatibility

packages/engine/src/remodel_engine/catalog/loader.py

@@ -58,22 +58,39 @@ def load_solar_profile(location_id: str) -> np.ndarray:
     Results are cached in-process — each location is read from disk only once.
     """
     path = _profile_path("solar", location_id)
+    data: np.ndarray | None = None
+
+    # First try: read header to check for %_generation column
+    with open(path, "r") as f:
+        header = f.readline().strip().lower()
+
+    if "%_generation" in header:
+        # New format: %_generation column (already percentage, 0-100)
+        cols = header.split(",")
         try:
-        # Format: col0=Hour_of_Year, col1=Date, col2=month, col3=day, col4=Hour_of_Day, col5=E_Grid_Kwp
-        data = np.loadtxt(path, delimiter=",", skiprows=1, usecols=5)  # E_Grid_Kwp
+            col_idx = cols.index("%_generation")
+            data = np.loadtxt(path, delimiter=",", skiprows=1, usecols=col_idx)
+            if data.max() > 1.0:
+                data = data / 100  # Convert % to fraction
+        except ValueError:
+            pass
+
+    if data is None:
+        try:
+            # Try old format: E_Grid_Kwp in kW (absolute for 1MW reference)
+            data = np.loadtxt(path, delimiter=",", skiprows=1, usecols=5)
+            if data.max() > 1.0:
+                data = data / 1000  # Convert kW to MW (for 1MW reference)
+                data = data / data.max()  # Normalize to [0,1]
         except:
             try:
-            # Format: hour,irradiance_norm
+                # Legacy: hour,irradiance_norm
                 data = np.loadtxt(path, delimiter=",", skiprows=1, usecols=1)
             except:
                 raise ValueError(f"Cannot parse solar profile {path}")
-    if data.shape[0] != 8760:
-        raise ValueError(f"Solar profile {path} has {data.shape[0]} rows, expected 8760")
 
-    # Normalize to [0, 1] if values are absolute (kWp)
-    max_val = data.max()
-    if max_val > 1.0:
-        data = data / max_val  # Normalize
+    if data is None or data.shape[0] != 8760:
+        raise ValueError(f"Solar profile {path} has {data.shape[0] if data is not None else 'None'} rows, expected 8760")
 
     return data.astype(np.float64)