Add an optional argument "hasnan" for qselect and partialsort

[r-devulap]

Handling NAN's slows down the algorithm. We make it an optional argument. Also fixes a bug in qselect and partialsort where NAN's were handled incorrectly. Improved benchmarks for float and doubles not containing NAN's:

Benchmark                                                                  Time             CPU      Time Old      Time New       CPU Old       CPU New
-------------------------------------------------------------------------------------------------------------------------------------------------------
[avx512_qselect<double> vs. avx512_qselect<double>]/10                  -0.1488         -0.1488          8922          7594          8928          7600
[avx512_qselect<double> vs. avx512_qselect<double>]/100                 -0.1542         -0.1540          8963          7582          8970          7588
[avx512_qselect<double> vs. avx512_qselect<double>]/1000                -0.1615         -0.1617          8664          7265          8672          7269
[avx512_qselect<double> vs. avx512_qselect<double>]/5000                -0.1472         -0.1472          9705          8276          9711          8281
[avx512_qselect<float> vs. avx512_qselect<float>]/10                  -0.1708         -0.1706          6049          5016          6052          5019
[avx512_qselect<float> vs. avx512_qselect<float>]/100                 -0.1691         -0.1690          6171          5128          6174          5131
[avx512_qselect<float> vs. avx512_qselect<float>]/1000                -0.1645         -0.1643          6229          5204          6231          5207
[avx512_qselect<float> vs. avx512_qselect<float>]/5000                -0.1586         -0.1581          6140          5166          6143          5172

Compared to stdnthelement:

Benchmark                                                                      Time             CPU      Time Old      Time New       CPU Old       CPU New
-----------------------------------------------------------------------------------------------------------------------------------------------------------
[stdpartialsort<float> vs. avx512_partial_qsort<float>]/10                  -0.2713         -0.2706          6952          5066          6953          5071
[stdpartialsort<float> vs. avx512_partial_qsort<float>]/100                 -0.7868         -0.7865         25051          5341         25053          5350
[stdpartialsort<float> vs. avx512_partial_qsort<float>]/1000                -0.9652         -0.9652        253841          8834        253840          8842
[stdpartialsort<float> vs. avx512_partial_qsort<float>]/5000                -0.9686         -0.9686        755803         23749        755779         23756
OVERALL_GEOMEAN                                                             -0.8858         -0.8857             0             0             0             0

Benchmark                                                                        Time             CPU      Time Old      Time New       CPU Old       CPU New
-------------------------------------------------------------------------------------------------------------------------------------------------------------
[stdpartialsort<double> vs. avx512_partial_qsort<double>]/10                  -0.0438         -0.0429          7958          7609          7957          7616
[stdpartialsort<double> vs. avx512_partial_qsort<double>]/100                 -0.6967         -0.6963         25841          7839         25843          7848
[stdpartialsort<double> vs. avx512_partial_qsort<double>]/1000                -0.9505         -0.9505        243181         12033        243190         12043
[stdpartialsort<double> vs. avx512_partial_qsort<double>]/5000                -0.9480         -0.9480        694575         36117        694556         36134

ping @mosullivan93

[mosullivan93]

Sorry, this notification got buried in my inbox until I saw the merge email. Thanks for patching up the partial functions, too.