Support add columns with all nulls

[wjones127]

User should be able to write:

dataset.add_columns([pa.field("new_col1", pa.int32()), pa.field("new_col2", pa.string())])

and it would add the new_col1 and new_col2 to the end of the schema. The values would all be null.

After finishing #3016, this can just be a metadata-only operation. Would just add to the schema, and not modify the data files.

[westonpace]

Is this feature the only motivation for #3016? All null data files should be quite small, regardless of the data type so they might not be so bad.

[wjones127]

Not at all. Motivation for #3016 is being able to create a schema and omit columns when inserting later.