Change use of compiler flags with GPU + C++11
Fix some unused parameter warnings in CUDA code, which were made visible
now that the warning flags propagate there.
Disable -Wnon-virtual-dtor, since C++-only flags seem to trip nvcc.
It is likely that this same warning is also given by cppcheck and/or the
clang static analyzer, so we should not lose much. We can still enable
it again if we implement partial propagation of flags to nvcc.
Refs #1390
Change-Id: I8489af3dcc3139884065abe2e5806d71992abd6c