Wrong function weight calculation

[cardillan]

Function weight calculation (used for cost/benefit analysis of some optimizations) is not computed correctly. Currently it is loosely based on the total number of function calls, but firstly, the #207 fix changes that in a convoluted way, and secondly, weight of the contexts containing the function calls is not considered (which is worse).

A proper calculation of function weights needs to be implemented. This will require much more detailed analysis, which might be also needed for correctly resolving function calls once types are implemented.