Fix UNION column alias aggregation

[kimjune01]

Summary

• When UNION combines queries with the same column alias, columns_aliases now aggregates all source columns into a list instead of silently overwriting
• Handles deduplication: repeated identical targets stay as a single value, different targets aggregate into a list

Fixes #401

Test plan

• pytest test/ -v: 270 tests pass
• New test test_union_column_aliases verifies {"M": ["tab1.A", "tab2.B"]} for UNION ALL with matching aliases
• Existing UNION tests unaffected

[collerek]

Good catch, thanks :)

Although we need some more fixes as this change will nest lists if the alias is referencing multiple columns, since this change changes the global resolution, not only the unions:

def test_union_alias_with_expression_targets():
      # Case A: scalar then list-target → produces a NESTED list
      q1 = """
      SELECT a AS x FROM t1
      UNION ALL
      SELECT b + c AS x FROM t2
      """
      p = Parser(q1)
      # PR currently returns {'x': ['a', ['b', 'c']]} — nested
      assert p.columns_aliases == {"x": ["a", "b", "c"]}

      # Case B: list then list-target → TypeError: unhashable type: 'UniqueList'
      q2 = """
      SELECT a + b AS x FROM t1
      UNION ALL
      SELECT c + d AS x FROM t2
      """
      p = Parser(q2)
      # PR currently raises TypeError: unhashable type: 'UniqueList.'
      assert p.columns_aliases == {"x": ["a", "b", "c", "d"]}

we need something like:

def add_alias(self, name: str, target: Any, clause: str) -> None:
      self.alias_names.append(name)
      if clause:
          self.alias_dict.setdefault(clause, UniqueList()).append(name)
      if target is None:
          return
      existing = self.alias_map.get(name, [])
      merged = UniqueList(existing if isinstance(existing, list) else [existing])
      merged.extend(target if isinstance(target, list) else [target])
      self.alias_map[name] = merged if len(merged) > 1 else merged[0]

[kimjune01]

Thanks for the careful review @collerek — both regressions reproduce exactly as you described:

• Case A (SELECT a AS x ... UNION ALL SELECT b + c AS x ...): produced {'x': ['a', ['b', 'c']]} (nested)
• Case B (SELECT a + b AS x ... UNION ALL SELECT c + d AS x ...): raised TypeError: cannot use 'sql_metadata.utils.UniqueList' as a set element from the target not in existing membership check

Pushed 4ca0cd6 adopting your suggested implementation verbatim. Both regression cases are now covered in test/test_unions.py::test_union_alias_with_expression_targets. Full suite: 271 passed.