Incorrect guess for Set.dimen for nested initialize kwarg + proposed fix

[mlangiu]

When guessing the dimension in sets.py: Set.init from the initialize kwarg, nesting is ignored.
However when accessing keys they are flattened. This results in incorrect guesses for cases such as this:

constraints = {c for c in itertools.product(['constrA', 'constrB'], range(5))}
vars = {v for v in itertools.product(['var1', 'var2', 'var3'], range(5))}
matrix_coefficients = {m for m in itertools.product(constraints, vars)}
m = ConcreteModel()
m.Matrix = Param(matrix_coefficients, default=0)

>>> ValueError: The value=('constrB', 4, 'var1', 1) does not have dimension=2, which is needed for set=Matrix_index

The code that is causing this sits in sets.py lines 691-695:

if type(self.initialize) is tuple:
    tmp = len(self.initialize)
elif type(self.initialize) is list and len(self.initialize) > 0 \
         and type(self.initialize[0]) is tuple:
    tmp = len(self.initialize[0])

One way to amended this is to flattening the entire self.initialize before checking with the current method or flattening the single element used for guessing, e.g. with pyutilib.misc.flatten.
If we're considering that nested containers other than tuple might occur, getattr seems more adequate than a direct len call.
Finally looking over init it seems like lines 691 and 692 are dead code, as tuples are already converted to lists in lines 682-685!

In summary it seems that changeing lines 691-695 to:

if type(self.initialize) is list and len(self.initialize) > 0:
    tmp = getattr(flatten(self.initialize[0]), 'len', 0)

should fix this behaviour.

[blnicho]

I'm going to add this to the list of known Set issues in #326