AMD GPU Fixes

[PaulMullowney]

This commit has 2 parts. The first uses the correct APIs to set 16KB of stack per device thread. It is this commit that enables more nalu reg tests to run. The 2nd commit removes the Experimental from the HIP namespacing.

Now, we see:
Start 1: fsiTurbineSurrogate
1/10 Test #1: fsiTurbineSurrogate .............. Passed 23.75 sec
Start 2: airfoilRANSEdgeNGPHypre.rst
2/10 Test #2: airfoilRANSEdgeNGPHypre.rst ...... Passed 38.38 sec
Start 3: ablNeutralNGPHypre
3/10 Test #3: ablNeutralNGPHypre ...............***Failed 21.60 sec
Start 4: ablNeutralNGPHypreSegregated
4/10 Test #4: ablNeutralNGPHypreSegregated .....***Failed 21.30 sec
Start 5: multiElemCylinder
5/10 Test #5: multiElemCylinder ................ Passed 26.71 sec
Start 6: VOFZalDisk
6/10 Test #6: VOFZalDisk .......................***Failed 19.30 sec
Start 7: airfoilSST_Gamma_Trans
7/10 Test #7: airfoilSST_Gamma_Trans ........... Passed 26.09 sec
Start 8: oversetRotCylNGPHypre
8/10 Test #8: oversetRotCylNGPHypre ............ Passed 26.66 sec
Start 9: convTaylorVortex
9/10 Test #9: convTaylorVortex .................***Failed 27.12 sec
Start 10: unitTestGPU
10/10 Test #10: unitTestGPU ...................... Passed 47.02 sec

The convTaylorVortex and VORZalDisk run to completion but fail based on diffs. The ablNeutralNGPHypre* tests seg fault at the beginning and require further investigation. It may be that we need more than 16KB per stack thread. This is under investigation.