Java: Improve finding best type for models and lifting.

[michaelnebel]

In this PR

• The logic for lifting models is improved (further explanation can be seen below).
• Fix a small issue with sink model exclusions (these were wrongly excluded based on summarized callables and neutrals with summary kind).
• Do some minor re-factoring.

The idea with the lifting logic is to make models as general as possible.
Assume that B <: A (B is a subtype of A or B is an implementation of A) and assume that we have
methods B.M and A.M where B.M is an override or implementation of A.M. In case the model generator finds a summary model S for B.M we lift the model to A.M (that is, we assume that A.M exhibits the same flow as B.M as this is perceived as a part of the override/implementation contract).

The old logic

• For lifting was dispersed throughout the code and it was slightly unsound (and unclear).
• Was not functional. The new implementation replaces superImpl with liftedImpl, which gives us a unique best callable to lift a model to (maybe the callable itself), if it is relevant to make a model for the callable.
• Lifted models to potential private/internal callables (this was identified by @owen-mc as a part of generating the java SDK models). Now models are only lifted to effectively public API endpoints.

Furthermore, we now only lift models to callables that are also in source. It seems a bit of stretch that if one implements an interface from an imported package and flow is identified for a method then we lift to the callable in the imported package (we could risk getting very odd models for stuff in the Java SDK).

As a test the models for the Java SDK were re-generated and it appears that

• Models for internal were not generated.
• Models for methods in eg. AbstractStringBuilder (which is package internal) will be only lifted to the relevant public endpoints.

[aschackmull]

LGTM