Move fast dual method from MICOM

[cdiener]

• description of feature/fix
• tests added/passed
• add an entry to the next release

This moves the fast dual methods from MICOM into optlang. This provides a quicker setup for primal/dual problems and multi-objective optimization.

Why?

• easier licensing
• method is not specific to MICOM
• code was originally adapted from convert_linear_problem_to_dual anyway

[Midnighter]

I'm a bit too far away from this at the moment, to be able to easily review. @KristianJensen can you take a look?

[Midnighter]

I'm not experienced with the dual problem representation, so I can't review for correctness at the moment. However, I do trust you on that part.

My take on the code is the following:

1. I know you prefer efficiency, but I think logically it would be clearer to first check the model for validity, i.e., loop through constraints and variables at the beginning and raise an exception before any modifications are made.
2. In the spirit of other functions, should this one be called add_fast_dual? Since it adds constraints and variables rather than returning a copy of the original model with modifications?
3. In addition to moving out validation, I would also introduce helper functions. One that handles a constraint and another that handles a variable. Perhaps a third that works on the objective. It would improve the structure and readability.

So in semi-pseudo code:

for constr in model.constraints:
    check_valid_constraint(constr)
for var in model.variables:
    check_valid_variable(var)

for constr in model.constraints:
    add_dual_constraint(constr)
for var in model.variables:
    add_dual_variable(var)

add_dual_objective(model.objective)

[Midnighter]

Either remove this, or lower the log level to debug, in my opinion.

Suggested change

	logger.info("adding dual variables")
	logger.debug("adding dual variables")

[Midnighter]

Here, I'm wondering if it is better to have it at warning level?

[Midnighter]

Suggested change

	logger.info("dual model has {} terms in objective".format(len(coefs)))
	logger.info("dual model has %d terms in objective", len(coefs))