Fix "direct" check in PipProvider.get_preference

[notatallshaw]

I noticed this while running tests from a user provided scenario in #12972. I'm going to have to think about how to add a test, but I'm confident that the direct is broken, possibly this happened when backjumping happened.

When has_information is true, candidate is always a tuple, so testing for candidate is not None is equivalent to testing has_information is true. This is clearly not the intent...

Looking at this tuple, the elements can either be None or an instance of Candidate which happens in both the case of ExplicitRequirement or RequiresPythonRequirement, and I assume this logic is trying to detect when a Candidate is based on ExplicitRequirement, but also being true for RequiresPythonRequirement doesn't functionally matter, as requires_python is tested for in front of direct.

Fixes #12975

[notatallshaw]

It seems there are no unit test cases related to ExplicitRequirement at all 🙁, so instead I'm going to mark ready for review and @uranusjr who I think added direct and @pradyunsg who got pip to support resolvelib's backjumping changes.

This bugfix will have a big impact on resolution preference, because "direct" is right near the top of preference and currently it's actually the same as "has_information", which I think is a bad preference choice, which I think means this should be a substantial improvement.

Happy to put in the work to add any extra stuff, but I need some direction.

[notatallshaw]

As I'm starting to get an proper unit test for get_preference together I'm coming around to two thoughts:

1. This never worked, before backjumping candidate was always a tuple
2. While I have an intuitive understanding of "direct", it's not clear to me from the code what this was even trying to test

[notatallshaw]

I started adding unit tests for each preference in get_preference, I have successfully added one for requires_python, but trying to add one for direct made me realize I don't understand what the code is intending to do, I wrote up an issue on this: #12975

[notatallshaw]

Closing this, new PR coming shortly with an actual bunch of unit tests.