Spark: Coerce shorts and bytes into ints in Parquet Writer

[shardulm94]

Fixes #10225

#9440 refactored the Spark Parquet writer to move to a visitor pattern. During this move, it removed some code required for coercion from bytes/shorts to int. This PR adds back the coercion code along with unit tests validating the change.

Note that even without coercion code, the tests succeed in Spark 3.5 but not in 3.3 and 3.4. It succeeds in 3.5 due to apache/spark#40734 which routes CTAS/RTAS through AppendData, in which case Spark adds its own projection to handle coercion here.

So the change is not strictly necessary for Spark 3.5 but I add it anyway to:

1. keep the code consistent between 3.3/3.4/3.5
2. future proofing in case Spark removes the auto-coercion at some point

[shardulm94]

cc: @Fokko @jkolash

@Fokko regarding your comment here, its actually different than what visiting IntLogicalTypeAnnotation achieves. The visitor will create a ByteWrite if the annotation of the Parquet logical type (based on Iceberg schema) says its a byte. In this case, the table schema says its an int (not byte), but the datatype of the Spark DF is byte and hence coercion is required.

[Fokko]

@shardulm94 This makes sense to me, thanks for fixing this 👍