Fix a potential segfault in PQ reader's number of rows per source calculation (#18906)

This PR fixes a potential segfault from illegal memory access (out of range vector index) in computation of number of rows read from each source in parquet reader. The segfault should manifest when the `num_rows` option leads us to not read from all data sources.

Consider this example: we have 5 files with 1000 rows each, we set our skip rows to 1500 and num rows to 1000. The `partial_sum_nrows_source` vector (at L725 in `reader_impl.cpp`) would look like this: [0, 500, 1000, 1000, 1000, 1000]. Here, computing the `end_iter` (end_row = 1000) with std::upper_bound will give us `std::end(..)` instead of `std::begin(..) + 2` which should lead to a segfault.

Authors:
  - Muhammad Haseeb (https://github.com/mhaseeb123)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Vyas Ramasubramani (https://github.com/vyasr)
  - Vukasin Milovanovic (https://github.com/vuule)
  - David Wendt (https://github.com/davidwendt)
  - Nghia Truong (https://github.com/ttnghia)

URL: #18906

Author: mhaseeb123

cpp/src/io/parquet/reader_impl.cpp

@@ -727,17 +727,14 @@ std::vector<size_t> reader::impl::calculate_output_num_rows_per_source(size_t co
   // Binary search start_row and end_row in exclusive_sum_num_rows_per_source vector
   auto const start_iter =
     std::upper_bound(partial_sum_nrows_source.cbegin(), partial_sum_nrows_source.cend(), start_row);
-  auto const end_iter =
-    (end_row == _file_itm_data.global_skip_rows + _file_itm_data.global_num_rows)
-      ? partial_sum_nrows_source.cend() - 1
-      : std::upper_bound(start_iter, partial_sum_nrows_source.cend(), end_row);
+  auto const end_iter = std::lower_bound(start_iter, partial_sum_nrows_source.cend(), end_row);
 
   // Compute the array offset index for both iterators
-  auto const start_idx = std::distance(partial_sum_nrows_source.cbegin(), start_iter);
-  auto const end_idx   = std::distance(partial_sum_nrows_source.cbegin(), end_iter);
-
-  CUDF_EXPECTS(start_idx <= end_idx,
-               "Encountered invalid source files indexes for output chunk row bounds");
+  auto const start_idx   = std::distance(partial_sum_nrows_source.cbegin(), start_iter);
+  auto const end_idx     = std::distance(partial_sum_nrows_source.cbegin(), end_iter);
+  auto const num_sources = static_cast<cudf::size_type>(partial_sum_nrows_source.size());
+  CUDF_EXPECTS(start_idx <= end_idx and start_idx < num_sources and end_idx < num_sources,
+               "Encountered out of range output table chunk row bounds");
 
   // If the entire chunk is from the same source file, then the count is simply num_rows
   if (start_idx == end_idx) {

cpp/tests/io/parquet_reader_test.cpp

@@ -2390,7 +2390,7 @@ TEST_F(ParquetReaderTest, NumRowsPerSource)
   // Read rows_to_read rows skipping rows_to_skip (> two sources) from ten data sources
   {
     auto constexpr rows_to_skip = 25'999;
-    auto constexpr rows_to_read = 47'232;
+    auto constexpr rows_to_read = 10'900;
 
     auto constexpr nsources = 10;
     std::vector<std::string> const datasources(nsources, filepath);