Fix radix test failure involving integer sign

[brian-kelley]

Now, the serial radix sorts enforce at compile time that the key is an unsigned integer.
This was really the assumption all along (that values were always
nonnegative). No longer adding a bias to the keys while sorting, since all the keys are at least 0.

This was causing the char radix sort unit test to fail in an OpenMP debug configuration - I think because of undefined overflow behavior when the chars were normalized from range (min, max) to range (0, max-min). This PR sorts only nonnegative chars and reinterprets as unsigned char, which is safe.

[brian-kelley]

Bowman PR: blas trsm tests failing due to #559, unrelated to these changes

#######################################################
PASSED TESTS
#######################################################
intel-16.4.258-Pthread-release build_time=726 run_time=1038
intel-16.4.258-Pthread_Serial-release build_time=1043 run_time=1992
intel-16.4.258-Serial-release build_time=696 run_time=942
intel-17.2.174-OpenMP-release build_time=883 run_time=561
intel-17.2.174-OpenMP_Serial-release build_time=1210 run_time=1481
intel-17.2.174-Pthread-release build_time=811 run_time=922
intel-17.2.174-Pthread_Serial-release build_time=1164 run_time=1806
intel-17.2.174-Serial-release build_time=775 run_time=900
#######################################################
FAILED TESTS
#######################################################
intel-18.2.199-OpenMP-release (test failed)
intel-18.2.199-OpenMP_Serial-release (test failed)
intel-18.2.199-Pthread-release (test failed)
intel-18.2.199-Pthread_Serial-release (test failed)
intel-18.2.199-Serial-release (test failed)

RIDE PR: failures are due to #344 (also unrelated)

#######################################################
PASSED TESTS
#######################################################
cuda-9.2.88-Cuda_OpenMP-release build_time=576 run_time=417
cuda-9.2.88-Cuda_Serial-release build_time=491 run_time=525
gcc-6.4.0-OpenMP_Serial-release build_time=204 run_time=395
gcc-7.2.0-OpenMP-release build_time=126 run_time=129
gcc-7.2.0-OpenMP_Serial-release build_time=216 run_time=361
gcc-7.2.0-Serial-release build_time=120 run_time=226
ibm-16.1.0-Serial-release build_time=494 run_time=398
#######################################################
FAILED TESTS
#######################################################
ibm-16.1.0-OpenMP-release (test failed)
ibm-16.1.0-OpenMP_Serial-release (test failed)

[srajama1]

Thanks @brian-kelley !

[srajama1]

Can we add documentation for sorting routines in the internal repo ?

[ndellingwood]

@brian-kelley did this get tested with Tpetra? I'm assuming this may impact spadd where we previously ran into an issue during promotion testing a couple weeks back?

[brian-kelley]

@ndellingwood Hmm, I didn't try it with Tpetra. But it doesn't call the sort functions directly so it should be fine. Spadd is currently the only place where these sorts are called, but I would like to update Tpetra::Details::sortCrsEntries to use them.