Stable for reference publication

Enables block-sparse matrix solving linking tfQMRgpu.so
Enables implicit C++/CUDA operators with the header-only tfQMRgpu::solve<action_t>