Change thrust::count_if call to raw kernel in strings split APIs

[davidwendt]

Description

Fixes calls to thrust::count_if in strings split APIs to better handle large strings.

Checklist

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

[karthikeyann]

LGTM 👍

[pmattione-nvidia]

What is the benefit of rolling your own kernel vs using count_if? In what way is it better?

[davidwendt]

What is the benefit of rolling your own kernel vs using count_if? In what way is it better?

Yes, good question. In this case, the thrust::count_if would technically not work because the begin/end iterators go beyond the max int32 type -- number of bytes in a strings column can now exceed max size_type. We've purposely compiled out some of the thrust calls that iterate beyond max int32 to help reduce the size of our binary. This affects only a couple of places including this one. We could certainly call count_if repeatedly but a custom kernel here is simple enough and will perform better.

[davidwendt]

/merge