Performance improvement in cudf::strings::join_strings for long strings

[davidwendt]

Description

Improves performance for longer strings with cudf::strings::join_strings() API.

The new implementation is inspired by recent changes in json/writer_impl.cu which uses the make_strings_children factor that accepts an iterator of string pointers to gather the results into a single column of data but does lose some performance for smallish-strings (less than 32 bytes). So the original code is refactored to work on smaller strings.
Also, a benchmark implementation is added in this PR to test against various sized input columns.
For longer strings, the performance improvement is up to 100x for very long strings (>512 bytes).

Reference #13048

Checklist

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

[davidwendt]

Some performance results from nvbench

|  row_width  |  num_rows  |   Ref Time |   Cmp Time |           Diff |   %Diff |
|-------------|------------|------------|------------|----------------|---------|
|     32      |    4096    |  69.168 us |  73.764 us |       4.596 us |   6.64% |
|     64      |    4096    |  75.721 us |  80.595 us |       4.874 us |   6.44% |
|     128     |    4096    | 100.534 us |  97.495 us |      -3.040 us |  -3.02% |
|     256     |    4096    | 168.141 us | 134.340 us |     -33.801 us | -20.10% |
|     512     |    4096    | 305.502 us |  73.091 us |    -232.411 us | -76.08% |
|    1024     |    4096    | 604.477 us |  76.163 us |    -528.314 us | -87.40% |
|     32      |   32768    |  72.283 us |  78.273 us |       5.989 us |   8.29% |
|     64      |   32768    |  94.454 us |  99.622 us |       5.169 us |   5.47% |
|     128     |   32768    | 144.520 us | 158.625 us |      14.105 us |   9.76% |
|     256     |   32768    | 263.790 us | 276.760 us |      12.970 us |   4.92% |
|     512     |   32768    | 479.498 us | 111.470 us |    -368.028 us | -76.75% |
|    1024     |   32768    | 918.149 us | 131.217 us |    -786.932 us | -85.71% |
|     32      |   262144   | 136.116 us | 143.007 us |       6.891 us |   5.06% |
|     64      |   262144   | 291.781 us | 299.467 us |       7.686 us |   2.63% |
|     128     |   262144   |   1.312 ms | 658.147 us |    -654.296 us | -49.85% |
|     256     |   262144   |   8.639 ms |   4.060 ms |   -4578.518 us | -53.00% |
|     512     |   262144   |  24.817 ms | 380.971 us |  -24435.926 us | -98.46% |
|    1024     |   262144   |  56.648 ms | 554.740 us |  -56093.582 us | -99.02% |
|     32      |  2097152   | 647.612 us | 676.242 us |      28.630 us |   4.42% |
|     64      |  2097152   |   1.810 ms |   1.843 ms |      32.039 us |   1.77% |
|     128     |  2097152   |  12.738 ms |   4.410 ms |   -8327.769 us | -65.38% |
|     256     |  2097152   |  75.984 ms |  26.904 ms |  -49080.225 us | -64.59% |
|     512     |  2097152   | 216.893 ms |   2.296 ms | -214597.167 us | -98.94% |
|     32      |  16777216  |   4.691 ms |   4.867 ms |     175.819 us |   3.75% |
|     64      |  16777216  |  13.816 ms |  13.999 ms |     183.007 us |   1.32% |

Smaller strings seem slightly slower but appears to be mostly noise since the code there did not change much. Also, all the differences are in microseconds.

[PointKernel]

Some non-blocking nits otherwise LGTM

[davidwendt]

/merge