ORC: Fix predicate pushdown for Not In and Not Equals

[shardulm94]

ORC uses SQL semantics for Search Arguments, so an expression like col != 1 will exclude rows where col is NULL along with rows where col = 1. In contrast, Iceberg's Expressions will keep rows with NULL values, so the equivalent ORC Search Argument for an Iceberg Expression col != x is col IS NULL OR col != x.

This PR fixes the issue of the ORC pushdown returning less rows than what Iceberg expects.

iceberg/data/src/test/java/org/apache/iceberg/data/TestMetricsRowGroupFilter.java

Lines 741 to 742 in 78e80d2

	// ORC-623: ORC seems to incorrectly skip a row group for a notIn(column, {X, ...}) predicate on a column which
	// has only 1 non-null value X but also has nulls

mentions that this might be a bug in ORC, but in fact its just a case of mismatched semantics. During conversion of Iceberg expression to ORC Search Arguments, we now take care of this semantic difference.

The wider discussion of SQL compatibility for Iceberg expressions is discussed on the dev list.

[rdblue]

+1

[rdblue]

Thanks for fixing this, @shardulm94!