Improve warm starting after model changes

[jajhall]

The simple rule of "hot start the unreduced LP if there is a basis" can be far from optimal

• If the basis is no longer square
• If steepest edge weights have to be computed
• If presolve reduces the LP greatly
• If "many" modifications have been made

Also, if a taboo basis change triggers model status kUnknown, then report what that basis change it was. Make sure that the cycling detection is cleared when re-starting, switching simplex algorithms, and removing perturbations - to avoid false positives. Tie this to #1979

Also, what triggers re-scaling?