fix(api): tolerate column aliases (procedure_name → procedure) + Spanish value mappings

El LLM de watsonx Orchestrate estaba alucinando nombres de columnas
(procedure_name en lugar de procedure) y los valores de procedimientos
en español sin traducir. Dos fixes:

1. Backend (benefits_api.py): mapas _COLUMN_ALIASES y _VALUE_ALIASES
   resuelven aliases comunes (procedure_name, name, plan, etc.) y
   traducciones ES→EN (radiografía → X Ray, resonancia → MRI, etc.)
   antes de aplicar el filtro. Mensajes de error ahora listan columnas
   válidas para que el agente se corrija solo.

2. OpenAPI spec (yaml + json): description de cada operación ahora
   enumera explícitamente las columnas válidas y los valores válidos
   del campo procedure, más una sección VALUE MAPPING ES→EN para que
   el LLM no tenga que adivinar.

app/benefits_api.py

@@ -27,6 +27,79 @@ _OPS = {
     "le": lambda s, v: s <= v,
 }
 
+_COLUMN_ALIASES = {
+    "procedure_name": "procedure",
+    "procedurename": "procedure",
+    "proc": "procedure",
+    "name": "member_name",
+    "member": "member_name",
+    "membername": "member_name",
+    "location_name": "location",
+    "provider": "location",
+    "facility": "location",
+    "rating": "facility_rating",
+    "facilityrating": "facility_rating",
+    "distance": "distance_miles",
+    "plan": "member_plan",
+    "memberplan": "member_plan",
+    "type": "procedure_type",
+    "proceduretype": "procedure_type",
+    "cost": "total_cost",
+    "totalcost": "total_cost",
+    "price": "total_cost",
+    "in_network_only": "in_network",
+    "innetwork": "in_network",
+}
+
+_VALUE_ALIASES = {
+    "xray": "X Ray",
+    "x-ray": "X Ray",
+    "radiografia": "X Ray",
+    "radiografía": "X Ray",
+    "radiografias": "X Ray",
+    "radiografías": "X Ray",
+    "resonancia": "MRI",
+    "resonancia magnetica": "MRI",
+    "resonancia magnética": "MRI",
+    "tomografia": "CT Scan",
+    "tomografía": "CT Scan",
+    "ct": "CT Scan",
+    "limpieza dental": "Dental Cleaning",
+    "limpieza": "Dental Cleaning",
+    "examen visual": "Vision Exam",
+    "examen de la vista": "Vision Exam",
+    "vision": "Vision Exam",
+    "chequeo anual": "Annual Physical Exam",
+    "examen anual": "Annual Physical Exam",
+    "fisico anual": "Annual Physical Exam",
+    "apendicectomia": "Appendectomy",
+    "apendicectomía": "Appendectomy",
+    "analisis de sangre": "Blood Test",
+    "análisis de sangre": "Blood Test",
+}
+
+
+def _normalize_column(col: str, valid_cols: list[str]) -> str | None:
+    if col is None:
+        return None
+    if col in valid_cols:
+        return col
+    lower = col.lower().replace(" ", "").replace("-", "_")
+    if lower in _COLUMN_ALIASES:
+        candidate = _COLUMN_ALIASES[lower]
+        if candidate in valid_cols:
+            return candidate
+    return None
+
+
+def _normalize_value(val):
+    if not isinstance(val, str):
+        return val
+    lower = val.lower().strip()
+    if lower in _VALUE_ALIASES:
+        return _VALUE_ALIASES[lower]
+    return val
+
 
 def _apply_filters_and_group(df: pd.DataFrame, filters_raw, group_by_raw) -> list[dict]:
     try:
@@ -35,21 +108,34 @@ def _apply_filters_and_group(df: pd.DataFrame, filters_raw, group_by_raw) -> lis
     except json.JSONDecodeError as exc:
         raise HTTPException(status_code=400, detail=f"Invalid JSON in filters or group_by: {exc}")
 
+    valid_cols = list(df.columns)
     result = df.copy()
     for f in filters:
-        col, op, val = f.get("column"), f.get("operator"), f.get("value")
-        if col not in result.columns:
-            raise HTTPException(status_code=400, detail=f"Unknown column: {col}")
+        raw_col, op, val = f.get("column"), f.get("operator"), f.get("value")
+        col = _normalize_column(raw_col, valid_cols)
+        if col is None:
+            raise HTTPException(
+                status_code=400,
+                detail=f"Unknown column: {raw_col}. Valid columns: {valid_cols}",
+            )
         if op not in _OPS:
-            raise HTTPException(status_code=400, detail=f"Unsupported operator: {op}")
-        result = result[_OPS[op](result[col], val)]
+            raise HTTPException(
+                status_code=400,
+                detail=f"Unsupported operator: {op}. Valid operators: {list(_OPS.keys())}",
+            )
+        norm_val = _normalize_value(val)
+        result = result[_OPS[op](result[col], norm_val)]
 
     if group_by:
-        missing = [c for c in group_by if c not in result.columns]
+        norm_group = [_normalize_column(c, valid_cols) for c in group_by]
+        missing = [orig for orig, n in zip(group_by, norm_group) if n is None]
         if missing:
-            raise HTTPException(status_code=400, detail=f"Unknown group_by columns: {missing}")
+            raise HTTPException(
+                status_code=400,
+                detail=f"Unknown group_by columns: {missing}. Valid columns: {valid_cols}",
+            )
         numeric_cols = result.select_dtypes(include="number").columns.tolist()
-        result = result.groupby(group_by)[numeric_cols].mean().reset_index()
+        result = result.groupby(norm_group)[numeric_cols].mean().reset_index()
 
     return result.to_dict(orient="records")