Fix(optimizer): don't merge CTEs with EXPLODE projections into outer scopes

[georgesittas]

Before:

>>> query = """
... SELECT Name,
...        FruitStruct.`$id`,
...        FruitStruct.value
...   FROM
...        (SELECT Name,
...                explode(Fruits) as FruitStruct
...           FROM fruits_table)
... """
>>> from sqlglot.optimizer import optimize
>>> print(optimize(query, dialect="spark").sql(dialect="spark", pretty=True))
SELECT
  `fruits_table`.`name` AS `name`,
  EXPLODE(`fruits_table`.`fruits`).`$id` AS `$id`,
  EXPLODE(`fruits_table`.`fruits`).`value` AS `value`
FROM `fruits_table` AS `fruits_table`

Spark doesn't like EXPLODE(...).`$id` apparently, so this PR aims to fix this issue:

spark-sql (default)> SELECT EXPLODE(ARRAY(CAST(STRUCT('fooo') AS STRUCT<a: STRING>))).a;
[FIELD_NOT_FOUND] No such struct field `a` in `col`.
spark-sql (default)> WITH cte(col) AS (SELECT EXPLODE(ARRAY(CAST(STRUCT('fooo') AS STRUCT<a: STRING>)))) SELECT col.a FROM cte;
a
fooo
Time taken: 3.967 seconds, Fetched 1 row(s)