Implement strings::repeat_strings

[ttnghia]

This PR implements strings::repeat_strings which repeats the given string(s) multiple times. In contrast with the existing API cudf::repeat that repeats the rows (copies one row into multiple rows), this new API repeats the string within each row of the given strings column (copies the content of each string multiple times into the output string). For example:

strs = ['aa', null, '',  'bbc']
out  = repeat_strings(strs, 3)
out is ['aaaaaa', null, '',  'bbcbbcbbc']

This implements cudf side API for NVIDIA/spark-rapids#68.

[harrism]

I think the name repeat_join is too ambiguous, because in libcudf, we use join to mean a very different thing: database join, meaning matching columns across different tables.

[ttnghia]

Yes, it was very difficult to choose a name for this API. You are welcomed to suggest a better name for this.

Basically, this API should be named repeat as in Spark. However, we already have cudf::repeat which just does repeating rows (for example, repeat(['a', 'bc'], 2) = ['a', 'a', 'bc', 'bc']. Since in strings namespace we have various join APIs that do similar tasks to this one (such as join_strings('a', 'bc') = 'abc'), I decided to call the new API repeat_join, which means "repeat, then join". I still feel this is not the best name, but I don't have a perfect name yet (another name would be join_repeated but I'm not sure if it's better than this one).

[harrism]

Strings are the only supported type, right? cudf::repeat_strings

[ttnghia]

Strings are the only supported type, right? cudf::repeat_strings

Yes, this API only supports strings. Should be cudf::strings::, and yes, repeat_strings is definitely better. Thanks Mark!!!

[davidwendt]

This is looking good.

[ttnghia]

@gpucibot merge

[robertmaynard]

+1 to CMake changes