Add compile flags to blt_add_library and executable macros

@cyrush says this should smooth out CUDA support. It also adds a nice feature though this could cause problems with adding compiler flags too liberally to mixed source libraries.

Locate the definitions of the blt_add_library and executable macros, then inspect how compiler flags are currently handled and how CUDA support is configured. Done means the macros support compile flags without applying them too broadly to mixed-source libraries, with the relevant build or test coverage updated.

Written by the indexing model from the issue text.