Support the "glp_simplex followed by glp_exact" mode that is mentioned in ​#18735 comment:7

For problems in which it works, it's much faster than pure glp_exact.

Component: numerical

Issue created by migration from https://trac.sagemath.org/ticket/20458