Assert for non-empty nulls

[razajafri]

Added assert check to ColumnView creation to ensure there are no non-empty nulls at the time of creation. This also contains changes to bypass this check by setting -da:ai.rapids.cudf.AssertEmptyNulls.

Description

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[razajafri]

@revans2

Would you mind taking a look at this when you have time. This addresses the issue of imported data having non-empty nulls. I just had to add another system property for the unit-tests to pass. PTAL

[razajafri]

There seem to be sync problems from the fork and the latest changes are not showing here.

[razajafri]

@revans2 @ttnghia

I think I have addressed all your concerns, PTAL

[razajafri]

CI failure has nothing to do with the changes in this PR.

[revans2]

Just a nit.

[razajafri]

@revans2 I wanted to see what you think about this.

Considering this is something that CUDF requires and assertions in Java aren't enabled by default, should we transition this to a hard check and throw an Exception instead of using an assert? We can do this at a later time too, I just wanted to know if we need a follow-on issue for this.

[revans2]

@revans2 I wanted to see what you think about this.

Considering this is something that CUDF requires and assertions in Java aren't enabled by default, should we transition this to a hard check and throw an Exception instead of using an assert? We can do this at a later time too, I just wanted to know if we need a follow-on issue for this.

We want the assertion off in production because checking if they are empty involves launching at least one kernel and is rather expensive. Having it be an assertion lets us verify that it is doing the right thing in tests, but we get the performance in production. It also lets us enable it in production if we think there is a problem. That said, in some cases where we know that we would get data corruption if we get a non-empty null, it is probably worth doing a check and then fixing the problem, at least until we feel comfortable enough that we have fixed all of the issues.

[razajafri]

rebuild

[razajafri]

/merge