fix(query): range sur colonnes non-numériques + blank/notBlank sur numérique/date (#41)

- `range` était ignoré silencieusement (TRUE, pas de filtre) sur les
  colonnes texte/date car seul `_numeric_to_sql` le gérait. Ajout du
  cas `range` dans la branche texte/date (BETWEEN, CAST VARCHAR pour
  les dates, comme les autres opérateurs de comparaison).
- `blank`/`notBlank` comparaient toujours à `''`, ce qui fait planter
  DuckDB (Conversion Error) sur les colonnes numériques/date. Le
  check est déplacé après le calcul de is_numeric/col_is_date : ces
  types utilisent IS [NOT] NULL sans comparaison à chaîne vide.

src/utils/query_ast.py

@@ -82,14 +82,18 @@ def _condition_to_sql(cond: Condition, schema: pl.Schema) -> tuple[str, list]:
     col_type = schema[col]
     quoted = f'"{col}"'
 
-    if cond.operator == "blank":
-        return f"({quoted} IS NULL OR {quoted} = '')", []
-    if cond.operator == "notBlank":
-        return f"({quoted} IS NOT NULL AND {quoted} <> '')", []
-
     is_numeric = col_type.is_numeric()
     col_is_date = col_type == pl.Date
 
+    if cond.operator == "blank":
+        if is_numeric or col_is_date:
+            return f"{quoted} IS NULL", []
+        return f"({quoted} IS NULL OR {quoted} = '')", []
+    if cond.operator == "notBlank":
+        if is_numeric or col_is_date:
+            return f"{quoted} IS NOT NULL", []
+        return f"({quoted} IS NOT NULL AND {quoted} <> '')", []
+
     if is_numeric:
         return _numeric_to_sql(cond, col_type, quoted)
 
@@ -106,6 +110,11 @@ def _condition_to_sql(cond: Condition, schema: pl.Schema) -> tuple[str, list]:
         return f"{quoted} IS NOT NULL AND {target} {op_map[cond.operator]} ?", [
             str(cond.value)
         ]
+    if cond.operator == "range":
+        return f"{quoted} IS NOT NULL AND {target} BETWEEN ? AND ?", [
+            str(cond.value),
+            str(cond.value2),
+        ]
     if cond.operator == "startsWith":
         return f"{quoted} ILIKE ?", [f"{cond.value}%"]
     if cond.operator == "endsWith":