fix(e6): map empty identifiers to empty_alias on the native executor

[tkaunlaky-e6]

Problem

Databricks treats `` (an empty name) as a normal column name. Checked on Spark 4.2, ANSI mode on and off:

• count(*) AS `` produces a column named '';
• that column can be referenced by `` everywhere: outer SELECT, qualified `s.```, through a CTE, `ORDER BY`, `GROUP BY`.

e6 cannot parse an empty identifier, so the transpiled AS "" failed with Encountered "\"". This is the "double-quoted string" variant in the TomTom native coverage sheet: 2 statements, one of which ran OK on Databricks.

Fix

With E6_EXECUTOR_TYPE=native, identifier_sql writes every empty identifier as empty_alias. Because the alias and every reference to it get the same name, they stay linked, e.g.

SELECT `` FROM (SELECT count(*) AS `` FROM t) s ORDER BY `` → SELECT empty_alias FROM (SELECT COUNT(*) AS empty_alias FROM t) AS s ORDER BY empty_alias

The java path is unchanged.

Behaviour vs Databricks

	Databricks (Spark 4.2)	e6 with this fix
Values, all shapes below	✓	same as Databricks
Column name	''	empty_alias
Two empty-named columns, then referenced by name	AMBIGUOUS_REFERENCE error	Column 'empty_alias' is ambiguous error

Shapes checked on both, with the same results: single and multiple empty aliases, empty alias on a plain column, subquery and CTE with SELECT *, outer SELECT ```, qualified s., CTE column referenced by empty name, `ORDER BY , GROUP BY ```, GROUP BYwithORDER BYordinals,UNION ALL`, and window functions.

Testing

• make check passes: ruff, ruff-format, mypy, and the full suite with both tokenizers (1013 tests each).
• test_empty_alias_native_executor covers the alias, a reference through a subquery plus ORDER BY, a normal alias (unchanged), and java (unchanged). 2 of its checks fail without the fix.
• The sheet's sample now gets past the parse error on a local e6 engine.