dynamically using LDL, LU or Cholesky solver in dense matrix solve to speed up calculations (relates #194)

REQUIRE now throws error instead of fatal, introduced INTERNAL_CHECK and changed some macro names to have a XERUS_ prefix

Added implementation for perform_trace() for sparse Tensors, added a test case for it and fixed a bug in fixed index handling in TN

TT::random no longer uses size_t to detect more-than-maximal ranks (and is more reliable in detecting larger than maximal ranks now)

custom allocator now enforces alignment even if it the default allocator returns non-aligned addresses