This closes FLINK-36854.

Consider the following simple schema:

[id INT NOT NULL COMMENT 'id_field' DEFAULT 17]

This would be sync to downstream sink databases that supports them intactly.

However, things would go wrong after being transformed like this:

transform:
  - projection: id, id AS id_alias, id + 1 AS id_new

Here, id_new column should carry no comments / default expressions obviously, since it has nothing to do with original id technically.

However, for id and id_alias column, it's reasonable to expect that comments and default expressions to be kept. Worse still, the nullability of id_alias will be lost since the previous half is regarded as a complex expression, and nullability of an expression's return type is uncertain.

So, currently the transform will yields the following schema:

[
  id INT NOT NULL,
  id_alias INT,
  id_new INT
]

With this PR, deduced schema will be something like:

[
  id INT NOT NULL COMMENT 'id_field' DEFAULT 17,
  id_alias INT NOT NULL COMMENT 'id_field' DEFAULT 17,
  id_new INT
]

Some extra parts have been changed in TransformParser#generateProjectionColumns since the previous version is a little cryptic.