Fix memory write error in get_list_child_to_list_row_mapping utility

[davidwendt]

Reference issue #8883 and depends on fixes in PR #8884

The get_list_child_to_list_row_mapping builds a map for rolling operation on a lists column. In the thrust::scatter call a map value includes the last offset which will always be out-of-bounds to given output vector. This output vector is used to build the resultant output map by calling thrust::inclusive_scan but the out-of-bounds offset value is not used -- which is why the utility does not fail. The fix in this PR simply allocates an extra row in the intermediate vector so the thrust::scatter will not write to out-of-bounds memory. Since the value is eventually ignored, it does not effect the result.

The code in this function was creating many temporary columns incorrectly using the passed in device_resource_manager variable mr. The code was corrected by changing these to be just device_uvector's instead making it more clear that these are internal temporary memory buffers. Further the code calling get_list_child_to_list_row_mapping utility is using the output as a temporary column and so this PR fixes the logic to correct the memory resource usage.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.10@3ea4b42). Click here to learn what that means.
The diff coverage is n/a.

❗ Current head 861572d differs from pull request most recent head 1ffaac9. Consider uploading reports for the commit 1ffaac9 to get more accurate results

@@               Coverage Diff               @@
##             branch-21.10    #8994   +/-   ##
===============================================
  Coverage                ?   10.71%           
===============================================
  Files                   ?      114           
  Lines                   ?    19090           
  Branches                ?        0           
===============================================
  Hits                    ?     2046           
  Misses                  ?    17044           
  Partials                ?        0           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3ea4b42...1ffaac9. Read the comment docs.

[robertmaynard]

CMake changes LGTM

[mythrocks]

Kudos for moving the non-template functions into the .cu. +1.

[karthikeyann]

Look good.

[davidwendt]

@gpucibot merge