[BUG] Parquet read fails when table created with more columns than file

[revans2]

Describe the bug
If you run a create table with more columns then the underlying parquet file has, the cpu version works and fills the column with nulls, but the gpu version blows up.

Steps/Code to reproduce bug

CREATE TABLE IF NOT EXISTS f_five_minutes (`five_minutes` TIMESTAMP, `day` DATE, `DateTime` TIMESTAMP)
USING PARQUET
OPTIONS (
path 's3://.../five_minutes'
)
PARTITIONED BY (day)

The parquet file itself only has five_minutes in it.

[revans2]

This also shows up when doing schema evolution with the 'mergeSchema' option.

An integration test has been added for this.

[revans2]

This was fixed as a part of 7294b1a