Add separator-on-null parameter to strings concatenate APIs

[davidwendt]

Closes #4728

This PR adds a new parameter to the cudf::strings::concatenate APIs to specify if separators should be added between null entries when the null-replacement (narep) parameter is valid. If the narep scalar is invalid (i.e. null itself) then the entire output row becomes null. If not, separators are added between each element. Examples:

s1 = ['a', 'b', null, 'dd', null]
s2 = ['A', null, 'CC', 'D', null]
concatenate( {s1, s2}, sep='+', narep=invalid ) -> ['a+A', null, null, 'dd+D', null]
concatenate( {s1, s2}, sep='+', narep='@' ) -> ['a+A', 'b+@', '@+CC', 'dd+D', '@+@']
concatenate( {s1, s2}, sep='+', narep='' ) -> ['a+A', 'b+', '+CC', 'dd+D', '+']

The new parameter is an enum separator_on_nulls which has YES or NO settings. The default parameter value will be YES to keep the current behavior as expected by Python cudf and for consistency with Pandas behavior.
Specifying NO here will suppress the separator with null elements (when narep is valid).

concatenate( {s1, s2}, sep='+', narep='', NO ) -> ['a+A', 'b', 'CC', 'dd+D', '']

This PR also changes the name of the cudf::strings::concatenate_list_elements API to cudf::strings::join_list_elements instead. The API pattern and behavior more mimic the cudf::strings::join_strings then the concatenate functions. Also, these are called by the Python join functions so the rename makes it more consistent with cudf.

This is a breaking change in order to make these APIs more consistent. Previously, the separators column version was returning nulls only for an all-null row. This has been changed to honor the separator_on_null parameter instead. Currently there was no Python cudf API calling this version. Only the rename required minor changes to the Cython layer.

The gtests were updated to reflect the new behavior. None of the pytests required any changes since the default parameter value matches the original behavior for those APIs that cudf actually calls.

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.06@3975f10). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-21.06    #8282   +/-   ##
===============================================
  Coverage                ?   82.88%           
===============================================
  Files                   ?      105           
  Lines                   ?    17874           
  Branches                ?        0           
===============================================
  Hits                    ?    14814           
  Misses                  ?     3060           
  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 3975f10...18b1e03. Read the comment docs.

[robertmaynard]

CMake changes look good to me

[kkraus14]

cmake / python lgtm

[tgravescs]

tested this latest patch out (along with pr 8285) and all my Spark tests pass so we have the desired null handling behavior.

[davidwendt]

@gpucibot merge