Improve key-value sort performance

[r-devulap]

1. AVX-512 32-bit key value sort now uses zmm registers exclusively
2. Added partition_avx512_unrolled function
3. Use get_pivot_blocks function in key-value sort.

Performance numbers:

Benchmark                                                                  Time             CPU      Time Old      Time New       CPU Old       CPU New
-------------------------------------------------------------------------------------------------------------------------------------------------------
[simdkvsort/random_1m vs. simdkvsort/random_1m]/uint64_t                -0.2048         -0.2048      21833844      17362725      21833373      17361179
[simdkvsort/random_1m vs. simdkvsort/random_1m]/int64_t                 -0.2099         -0.2099      21922325      17320157      21919807      17319054
[simdkvsort/random_1m vs. simdkvsort/random_1m]/double                  -0.2257         -0.2258      20498044      15871414      20496653      15869322
[simdkvsort/random_1m vs. simdkvsort/random_1m]/uint32_t                -0.5045         -0.5045      16658607       8254529      16657580       8253890
[simdkvsort/random_1m vs. simdkvsort/random_1m]/int32_t                 -0.5009         -0.5009      16618624       8293955      16617563       8293402
[simdkvsort/random_1m vs. simdkvsort/random_1m]/float                   -0.4644         -0.4645      16973247       9090381      16972334       9089449
OVERALL_GEOMEAN                                                         -0.3668         -0.3669             0             0             0             0