Readability cleanup in AppPrediction comparator.

(i.e., in AppPredictionServiceResolverComparator)

This is believed to be a pure refactoring; no behavioral changes.

This is relatively simple compared to the ongoing cleanup of the
other comparator (e.g. in ag/17697713) because the AppPrediction
comparator already used simple immutable model data. This should
be more clear after this CL, which shows that the updated model
instance is derived from the sorted list we receive from the
service, along with immutable data that was provided when the
AppPredictionServiceResolverComparator was initialized.

The only parts that remain in the legacy ...ResolverComparator
code are responsible either for fetching the model data from the
remote service (which uses a somewhat sloppy/ad-hoc concurrency
model that should be cleaned up in a future CL as part of removing
the AbstractResolverComparator's Handler design), or for implementing
a fallback to use a ResolverRankerServiceResolverComparator if
there's a problem with the AppPredictionService (which should also
be cleaned up in a future CL, since the AppPredictionService
comparator really doesn't need to take on ownership responsibilities
for some completely different model).

Note that this CL also removes the bookkeeping of mTargetScores
(including the `checkAppTargetRankValid()` method) because it
wasn't used for anything.

Test: atest ResolverActivityTest ChooserActivityTest
Bug: 227486788
Change-Id: I419a73bf094c14276b0c5791edacaa27797b215c