Rework logic in cudf::strings::split_record to improve performance

[davidwendt]

Description

Updates the cudf::strings::split_record logic to match the more optimized code in cudf::strings:split.
The optimized code performs much better for longer strings (>64 bytes) by parallelizing over the character bytes to find delimiters before determining split tokens.
This led to refactoring the code so it both APIs can share the optimized code.
Also fixes a bug found when using overlapped delimiters.
Additional tests were added for multi-byte delimiters which can overlap and span multiple adjacent strings.

Closes #12694

Checklist

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

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (branch-23.04@ec8704a). Click here to learn what that means.
Patch has no changes to coverable lines.

❗ Current head d7dcb2a differs from pull request most recent head 524e038. Consider uploading reports for the commit 524e038 to get more accurate results

Additional details and impacted files

@@               Coverage Diff               @@
##             branch-23.04   #12729   +/-   ##
===============================================
  Coverage                ?   85.85%           
===============================================
  Files                   ?      158           
  Lines                   ?    25204           
  Branches                ?        0           
===============================================
  Hits                    ?    21638           
  Misses                  ?     3566           
  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.

[davidwendt]

Performance numbers for cudf::strings::split_record

/record/4096/32          0.186         0.262      0.71x
/record/4096/64          0.254         0.270      0.94x
/record/4096/128         0.408         0.281      1.45x
/record/4096/256         0.845         0.307      2.75x
/record/4096/512          2.25         0.280      8.04x
/record/4096/1024         7.89         0.307     25.70x
/record/4096/2048         29.3         0.376     77.93x
/record/4096/4096          115         0.498    230.92x
/record/4096/8192          464         0.752    617.02x
/record/32768/32         0.204         0.274      0.74x
/record/32768/64         0.279         0.297      0.94x
/record/32768/128        0.506         0.349      1.45x
/record/32768/256         1.05         0.444      2.36x
/record/32768/512         2.71         0.413      6.56x
/record/32768/1024        8.36         0.561     14.90x
/record/32768/2048        36.5         0.861     42.39x
/record/32768/4096         164          1.46    112.33x
/record/32768/8192         706          2.70    261.48x
/record/262144/32        0.328         0.419      0.78x
/record/262144/64        0.576         0.581      0.99x
/record/262144/128        1.47          1.16      1.27x
/record/262144/256        5.65          3.62      1.56x
/record/262144/512        12.3          1.51      8.15x
/record/262144/1024       48.3          2.71     17.82x
/record/262144/2048        216          5.59     38.64x
/record/262144/4096       1066          11.8     90.34x
/record/2097152/32        1.24          1.47      0.84x
/record/2097152/64        2.85          2.73      1.04x
/record/2097152/128       8.27          6.51      1.27x
/record/2097152/256       31.9          23.1      1.38x
/record/2097152/512       55.6          10.5      5.30x
/record/16777216/32       8.79          10.1      0.87x
/record/16777216/64       21.3          20.2      1.05x

[PointKernel]

LGTM with some nits (not necessarily a change request though)

[davidwendt]

/merge