[BUG] String corruption when skipping rows

[etseidl]

Spark-rapids uncovered an error in string processing NVIDIA/spark-rapids#9948. The issue arises when lists are present, num_rows/skip_rows is utilized, and the string column contains rows that begin and end in different pages. In this particular case, the code used to calculate page start/end value indices incorrectly set the start value to 0, resulting in an overestimation of the string buffer size, and the introduction of uninitialized data to the resulting column.

This condition is hard to test for because libcudf will always start a page at a new row boundary.