Add method to solve the LP corresponding to discrete variables fixed at the optimal solution of a MIP

[jajhall]

In other words, add the equivalent of GRBfixmodel to HiGHS.

Motivated by #2556 , but something that was planned when discovering that some folk want dual information and/or a basis at the solution of a MIP.

[peno64]

Note that semi-continious variables are a special case. For integer/binary variables, the fixed model is fixing the integer/variable value to its obtained (integer) value and remove the integer attribute.
However for semi-cont variables you have to look at the two possible states the variable can obtain. If its value becomes 0, then also a min/max of 0 must be put on this variable (ie fixed to 0). However if the value is non-zero and thus >= minimum, the variable may not be fixed on its value but instead the min/max on this variable must be set. Otherwise wrong dual information can be returned.

[peno64]

Except if the variable is semi-cont and integer in which I guess it must be treated as integer and this fixed on its obtained value.

[jajhall]

Indeed, this is what I've done with semi-continous and semi-integer variables

[jajhall]

Closed by #2559