Tolerance bug for double type join on columns

[Tom-Newton]

When leftPitKey or rightPitKey id a double type the interpreted code path produces wrong results.

The problem is in this code

tolerance > 0 && (
              leftRowPITKey
                .getLong(0) - rightRowPITKey.getLong(0) > tolerance
            )

spark-pit/scala/src/main/scala/execution/PITJoinExec.scala

Line 726 in 452dee7

.getLong(0) - rightRowPITKey.getLong(0) > tolerance