tests/preconditioned_conjugate_gradient_method_tests.m: make reliable?
This was ultimately made reliable by the new "condmax" parameter to
random_positive_definite_matrix(), but there are some additional
tweaks afterwards:
* Reorder / rename some stuff to improve readability
* Use a higher condmax than default, since we are preconditioning
* Pass the full "M" matrix to pcg(), to more closely match what
our own PCG function is doing.