Improving the purity analysis

[errt]

Taken from Bitbucket#157

Can we do something about the following issue (from the ASE paper): "In contrast to Benton and Fischer, we treat methods reading the receiver's mutable state as \property{side-effect free} instead of \property{externally pure}. This enables us to classify methods with calls on non-confined receiver objects as \property{side-effect free} rather than \property{impure}. The drawback is that a caller that invokes such methods on a confined receiver object can not be \property{pure} anymore."