Performance improvement for cudf::string_view::find functions

[davidwendt]

Description

Improves performance of calls to cudf::string_view::find functions for longer strings by minimizing character and byte counting required for resolving the input parameters. The input values are in character positions/counts so range checks do not include bytes that may be in partial characters. Converting these to bytes requires some counting usually from the beginning of the string for any given character position. The code has been reworked to minimizing this counting as much as possible. This improves performance overall by 10% but greatly improves performance for long strings by up to 3x.

Closes #12445

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]

For cudf::strings::find

rows	width	baseline (ms)	rework (ms)	x
4096	32	0.041	0.038	1.1
4096	128	0.075	0.060	1.3
4096	512	0.234	0.169	1.4
4096	2048	0.948	0.465	2.0
4096	8192	4.20	1.40	3.0
32768	32	0.044	0.040	1.1
32768	128	0.081	0.065	1.2
32768	512	0.254	0.197	1.3
32768	2048	1.32	0.523	2.5
32768	8192	4.73	1.48	3.2
262144	32	0.076	0.061	1.2
262144	128	0.215	0.177	1.2
262144	512	3.80	1.40	2.7
262144	2048	21.3	5.70	3.7
2097152	32	0.325	0.215	1.5
2097152	128	1.24	0.944	1.3
2097152	512	30.6	10.0	3.1
16777216	32	2.31	1.44	1.6

For cudf::strings::find_multiple

rows	width	baseline (ms)	rework (ms)	x
4096	32	0.048	0.045	1.1
4096	128	0.094	0.073	1.3
4096	512	0.255	0.195	1.3
4096	2048	0.737	0.433	1.7
4096	8192	3.36	1.14	2.9
32768	32	0.053	0.048	1.1
32768	128	0.104	0.083	1.3
32768	512	0.302	0.230	1.3
32768	2048	1.19	0.585	2.0
32768	8192	4.13	1.51	2.7
262144	32	0.123	0.092	1.3
262144	128	0.359	0.239	1.5
262144	512	1.55	0.993	1.6
262144	2048	5.66	2.70	2.1
2097152	32	0.695	0.439	1.6
2097152	128	2.27	1.37	1.7
2097152	512	9.72	6.30	1.5
16777216	32	5.28	3.21	1.6

[bdice]

Looks good to me. Nice optimization!

[davidwendt]

/merge