* Tried to fix the dense triangular solvers, yet my AMD driver is buggy...

* Single-threaded backend now also supports sparse matrices.
* Also cleaned-up the sparse matrix handling, now there is only sparse_matrix_operations.hpp instead of compressed_matrix_operations.hpp, coordinate_matrix_operations.hpp, etc. As a side-effect, non-CSR matrices now also support all the operations on CSR matrices. :-)