Support manifest lists and use the publish RID to pick the most relevant manifest from the list

[baronfel]

The basic idea is:

• prefer getting manifest lists from registries if they are available
• if none is available, use the existing manifest code
• if a list is available, cross-reference the items on that list with the publish RID to determine which image in the list should be the 'real' base image, the continue the rest of the process per normal.

This does not enable creating manifest lists for a project, just enables users to trust that the -r parameter to publish will do the right thing.

TODO:

• Dedicated tests, positive and negative for the rid-probing mechanism
• Parameterized end-to-end tests of the same
• Error messages and codes for the new thrown exceptions

[baronfel]

This is ready for review.

We read manifest lists, convert the available platform-specific manifests to equivalent RIDs and extract those slices of the RID graph, and then find the most-applicable base image for the RID the user specified for the build. If there is no match, we emit an error with the valid values. It works for linux-x64, linux-arm, and linux-amr64 on my local testing with a local Docker daemon.

The only thing I don't like is the hard-coded path for the RuntimeGraphFilePath in the tests - right now we pin to the 7.0.100 SDK but ideally we'd be able to get that from the runtime environment somehow.

[baronfel]

@vijayrkn as a side effect of this change, the VS side of the integration that forces linux-x64 could be lifted. We'll have to talk about the exact mechanism though - perhaps package version is enough to feature-gate that logic?

[clarkezone]

Is this available to test in a build?

[baronfel]

Yes! Version 0.3.2 is available on NuGet right now, and the documentation in this repo has been updated to match it. We've also got release notes up here. Try it out and let us know what you think. I'll be working with our friends in the docs org to update our pages on learn.microsoft.com over this week.