Parameter and Variable shapes

[pckroon]

This is related to #175.
Currently, Parameters must be scalars, and Variables can be any shape (scalar, vector, matrix, N-dimensional tensor, ...). This distinction stems from the scipy minimize interface, which demands all parameters to be scalars. This distinction between Variables and Parameters is rather annoying and limiting at times, however.

For Variables, the user provides the values and the shapes. For Parameters, we provide the values (initial values sometimes come from users, usually from us; scipy does the minimization, and we return those found values to the user). Users should provide their shape.

One way of approaching this would be to become agnostic to the underlying sympy objects, and accept e.g. Indexed, Symbol, Matrix, etc.