Shrink size of block GS test matrix

[brian-kelley]

so that tests don't time out. Decreased number of rows by 4x, and entries per row by a third.
In a build with -g, non optimized, and Kokkos
debug, this reduced the time to run all block GS tests from 43s to 3.2s.
I don't think the test coverage is any worse, and anyway block GS is not used anywhere currently.

#######################################################
PASSED TESTS
#######################################################
clang-8.0-Cuda_OpenMP-release build_time=712 run_time=438
clang-8.0-Pthread_Serial-release build_time=223 run_time=128
clang-9.0.0-Pthread-release build_time=131 run_time=58
clang-9.0.0-Serial-release build_time=136 run_time=51
cuda-10.1-Cuda_OpenMP-release build_time=976 run_time=149
cuda-9.2-Cuda_Serial-release build_time=899 run_time=165
gcc-4.8.4-OpenMP-release build_time=144 run_time=52
gcc-7.3.0-OpenMP-release build_time=149 run_time=51
gcc-7.3.0-Pthread-release build_time=118 run_time=57
gcc-8.3.0-Serial-release build_time=184 run_time=54
gcc-9.1-OpenMP-release build_time=230 run_time=132
gcc-9.1-Serial-release build_time=208 run_time=53
intel-17.0.1-Serial-release build_time=339 run_time=63
intel-18.0.5-OpenMP-release build_time=405 run_time=1819
intel-19.0.5-Pthread-release build_time=398 run_time=60

[ndellingwood]

Thanks @brian-kelley !