[REVIEW]Optimize pg.add_data for vector properties

[VibhuJawa]

This PR fixes #2991

Optimization around 4000x . See benchmarks.

On this PR:

------------------------------------------------------------------ benchmark: 12 tests -------------------------------------------------------------------
Name (time in ms, mem in bytes)                 Mean                   GPU mem            GPU Leaked mem            Rounds            GPU Rounds          
----------------------------------------------------------------------------------------------------------------------------------------------------------
bench_add_vector_features[32-10000]          21.3116 (1.0)           4,050,032 (1.0)                   0 (1.0)           1           1
bench_add_vector_features[32-100000]         23.9746 (1.12)         40,500,032 (10.00)                 0 (1.0)           1           1
bench_add_vector_features[64-100000]         32.7488 (1.54)         78,900,032 (19.48)                 0 (1.0)           1           1
bench_add_vector_features[64-10000]          33.9622 (1.59)          7,890,032 (1.95)                  0 (1.0)           1           1
bench_add_vector_features[128-10000]         46.1369 (2.16)         15,570,032 (3.84)                  0 (1.0)           1           1
bench_add_vector_features[32-1000000]        51.0027 (2.39)        405,000,032 (100.00)                0 (1.0)           1           1
bench_add_vector_features[128-100000]        56.5531 (2.65)        155,700,032 (38.44)                 0 (1.0)           1           1
bench_add_vector_features[32-10000000]      102.4596 (4.81)      4,050,000,032 (999.99)                0 (1.0)           1           1
bench_add_vector_features[64-1000000]       104.9687 (4.93)        789,000,032 (194.81)                0 (1.0)           1           1
bench_add_vector_features[64-10000000]      181.3918 (8.51)      7,890,000,032 (>1000.0)               0 (1.0)           1           1
bench_add_vector_features[128-1000000]      455.6243 (21.38)     1,557,000,032 (384.44)                0 (1.0)           1           1
bench_add_vector_features[128-10000000]     566.9107 (26.60)    15,570,000,032 (>1000.0)               0 (1.0)           1           1
----------------------------------------------------------------------------------------------------------------------------------------------------------

On Mainline:

---------------------------------------------------------------- benchmark: 2 tests ---------------------------------------------------------------
Name (time in s, mem in bytes)               Mean               GPU mem            GPU Leaked mem            Rounds            GPU Rounds          
---------------------------------------------------------------------------------------------------------------------------------------------------
bench_add_vector_features[32-10000]        5.6117 (1.0)       6,489,888 (1.0)                   0 (1.0)           1           1
bench_add_vector_features[32-100000]     102.9704 (18.35)    64,899,888 (10.00)                 0 (1.0)           1           1
...
...
...
...
---------------------------------------------------------------------------------------------------------------------------------------------------

[VibhuJawa]

rerun tests

[rlratzel]

This is great, thanks for adding the benchmark too. I had one question below:

[rlratzel]

Is this intended to be a public API, or is there a risk cuDF could change it and break us?

[VibhuJawa]

There is a risk that cuDF could change it later on but I could not come up with something without it.

https://github.com/search?q=repo%3Arapidsai%2Fcudf+bools_to_mask+language%3APython&type=code&l=Python

[rlratzel]

@gpucibot merge