Fix new resolver when using transitive pinning to resolve subgraphs correctly

[nkolev92]

Bug

Fixes: NuGet/Home#13938

Description

This fixes the new dependency resolver when transitive pinning is enabled and a package subgraph happens to contain versions that should or should not be eclipsed.

The way eclipsing works is for each subgraph, the first one wins:

Project1
└── A 1.0.0
    ├── B 1.0.0  
    └── C 1.0.0
        └── B 2.0.0

In this case, B 2.0.0 is part of the graph of A which already depends on B 1.0.0 so B 1.0.0 is chosen.

However, if a higher version comes in through a different subgraph, a different version of B could win:

Project1
├── A 1.0.0
│   ├── B 1.0.0  
│   └── C 1.0.0
│       └── B 2.0.0
└── F 1.0.0
    └── B 3.0.0

In this case, B 3.0.0 is not in the subgraph of A and its version eclipses B 1.0.0.

Transitive pinning is supposed to be syntactic sugar for a direct dependency and so if you pinned C 1.0.0 in the first example, the resolver treats the graph as:

Project1
├── A 1.0.0
│   └── B 1.0.0  
└── C 1.0.0
    └── B 2.0.0

Now B 2.0.0 is not part of the subgraph of A and the resolver should pick B 2.0.0. However, the current implementation of the new resolver only looks to see if the proposed B 2.0.0 and the already chosen B 1.0.0 have the common ancestors. This leads to the new resolver resolving the wrong graph and logging downgrade warnings as described in NuGet/Home#13938

The fix is to modify the paths for each item if it is pinned to correctly assess if a package should eclipse another. But when the paths are modified, the resolver must also keep track of the original parent in order to correctly create downgrade warnings if one exists.

PR Checklist

• Meaningful title, helpful description and a linked NuGet/Home issue
• Added tests
• Link to an issue or pull request to update docs if this PR changes settings, environment variables, new feature, etc.

[jgonz120]

Seems fine! I definitely don't know much about this area, that method is massive though. A refactor would definitely help make it easier to understand.

[nkolev92]

The changes in DependencyGraphResolver look correct.

Still not 100% convinced RestoreCommand_WithPackageDrivenDowngradeAndTransitivePinning_ElevatesTransitiveToDirect_AndDoesNotRaiseWarning is the best solution, despite what the old algorithm did.

I did want to do add some IncludeAssets/ExcludeAssets tests to make sure they're not broken with this change.