Add PackageFinder.make_candidate_evaluator()

[cjerdonek]

This is a follow-on to the refactoring in PR #6684.

The main thing this PR does is add a make_candidate_evaluator() method to the PackageFinder class to parallel the make_link_evaluator() method. This sets things up to allow for per-requirement CandidateEvaluator objects, which will be useful for later PR's.

This PR also does the following related things:

1. Add a CandidatePreferences class to encapsulate most of the options needed for CandidateEvaluator. (A later PR can add a parallel LinkPreferences class for LinkEvaluator.) Both of these will help to cut down the number of arguments / attributes needed for the PackageFinder class. It also helps the reader to conceptualize what "package-finding" stage the various options / attributes are needed for.
2. Refactors out from CandidateEvaluator.make_found_candidates() a separately testable CandidateEvaluator.get_applicable_candidates() method. This isolates to one method the logic for filtering the "non-applicable" candidates out from the list of all candidates. This PR also adds a unit test of this method.

[chrahunt]

Have we considered going the direction of the caller passing in a CandidateEvaluator factory instead of bundled preferences/options used by PackageFinder itself to instantiate CandidateEvaluator(s)? We would still get to the eventual goal of per-requirement CandidateEvaluator objects, and at the same time reduce some of the boilerplate related to translating arguments between different structures.

[chrahunt]

Out of curiosity, what is the style convention for the offset of these type declaration comments? In some places it is aligned with the others in a declaration, in others there are chunks of aligned comments, and in others they are all just 2 spaces from their corresponding parameters.

[cjerdonek]

I don't know of one. I've always been accustomed to two spaces for inline comments, but some people started aligning them. I think it's overboard to always align (e.g. having to change every argument line just because of a change in one line), so I sometimes align to adjacent lines when convenient and it improves the readability / how it looks.

[cjerdonek]

Have we considered going the direction of the caller passing in a CandidateEvaluator factory instead of bundled preferences/options used by PackageFinder itself to instantiate CandidateEvaluator(s)?

For now I just wanted to keep things unclever while also keeping things flexible and not making changes that would require unneeded changes to test code. I did briefly consider CandidateEvaluatorFactory but thought that introduced too much abstraction for not so much benefit. Things are still changing, so we can always revisit if we find that something will provide advantages.

[chrahunt]

LGTM.

[cjerdonek]

Thanks for the review and helpful comments, @chrahunt!

[cjerdonek]

(Merging this now so I can work on the follow-on PR.)