version 4

• Slightly optimized (8-12% faster for D10):
  • used continuous memory allocated at single step and distributed manually,
  • some of copy operations of matrices are sequential.
• Structure for facet is aggregate.
• Main and sole build system is CMake.