Supernode-based SpTrsv updates

[iyamazaki]

This PR

• adds "partitioned-inverse" option based on SpMV through Cholmod interface

• merge some changes for improving performance

[iyamazaki]

Spot-check on White:

../scripts/cm_test_all_sandia --spot-check --arch=Power8,Pascal60 --num=1

#######################################################
PASSED TESTS
#######################################################
cuda-10.1.105-Cuda_OpenMP-release build_time=590 run_time=168
cuda-10.1.105-Cuda_Serial-release build_time=657 run_time=185
cuda-9.2.88-Cuda_OpenMP-release build_time=613 run_time=183
cuda-9.2.88-Cuda_Serial-release build_time=591 run_time=202
gcc-6.4.0-OpenMP_Serial-release build_time=235 run_time=145
gcc-7.2.0-OpenMP-release build_time=146 run_time=65
gcc-7.2.0-OpenMP_Serial-release build_time=206 run_time=142
gcc-7.2.0-Serial-release build_time=136 run_time=71
ibm-16.1.1-Serial-release build_time=973 run_time=74

../scripts/cm_test_all_sandia --spot-check-tpls --arch=Power8,Pascal60 --cxxflags-extra='-DKOKKOSKERNELS_ENABLE_SUPERNODAL_SPTRSV=ON' --num=1

#######################################################
PASSED TESTS
#######################################################
cuda-10.1.105-Cuda_Serial-release build_time=731 run_time=184
cuda-9.2.88-Cuda_OpenMP-release build_time=719 run_time=188
gcc-7.2.0-OpenMP-release build_time=165 run_time=63
gcc-7.2.0-Serial-release build_time=153 run_time=72
gcc-7.4.0-OpenMP-release build_time=178 run_time=60

[ndellingwood]

LGTM, passes spot-checks, thanks @iyamazaki !