Performance tuning the sampling primitive for multi-node multi-GPU systems.

[seunghwak]

• Update groupby code in multi-GPU communication to use atomics based partitioning instead of sort based partitioning (with atomics performance updates in recent NVIDIA GPUs, now the atomics based approach is significantly faster than the sorting based approach if the number of groups is not excessive).
• In random index generation, add an additional code to handle high-degree vertices with with_replacement = false.

[codecov-commenter]

Codecov Report

❗ No coverage uploaded for pull request base (branch-23.04@a0d964d). Click here to learn what that means.
Patch has no changes to coverable lines.

Additional details and impacted files

@@               Coverage Diff               @@
##             branch-23.04    #3169   +/-   ##
===============================================
  Coverage                ?   56.26%           
===============================================
  Files                   ?      153           
  Lines                   ?     9658           
  Branches                ?        0           
===============================================
  Hits                    ?     5434           
  Misses                  ?     4224           
  Partials                ?        0           

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[ajschmidt8]

Removing ops-codeowners from the required reviews since it doesn't seem there are any file changes that we're responsible for. Feel free to add us back if necessary.

[rlratzel]

(approving for python-codeowners to unblock merging, but didn't review any code here - assuming python-codeowners added from a file change no longer in this PR)

[seunghwak]

(approving for python-codeowners to unblock merging, but didn't review any code here - assuming python-codeowners added from a file change no longer in this PR)

Yes, this is due to branch re-targeting.

[ChuckHastings]

/merge