Suggesting we rewrite (parts of) imgutil

[apostasie]

What is the problem you're trying to solve

In light of #3516 it is clear there are issues with code that likely pre-dates the introduction of multi-architecture.

There is also a certain amount of duplication - a lot of what pkg/cmd/image/list.go does should really be provided by imgutil.

Also, imgutil is making assumptions that the current native platform version of the image is necessarily available - or that other platforms variants simply do not matter (especially wrt labels).

Finally, the imagewalker (see #3502) might be part of the problem - along with the added abstraction layers (EnsuredImage) on top / mixed with the already crowded and confusing containerd API (Image aka provides the model for how containerd views container images. vs. Image aka describes an image used by containers - client.ImageService().List(ctx) vs. client.ListImages(ctx) 🤦‍♂️).

Describe the solution you'd like

Suggesting someone motivated starts from scratch:

• cleanup the confusion between referenceutil and distributionref, and come-up with one simple abstraction to represent image references (and carry that around instead of re-parsing the rawref over and over again)
• review all the places where we are currently manipulating Images
• list what they need
• come-up with a multi-platform-aware abstraction for "NerdImage" to cover all of these use-cases (maybe that is an evolution of EnsuredImage) and seal the implementation away so that we stop shooting ourselves in the foot with containerd types
• re-evaluate all helpers inside imgutil against use cases - simplify them, make them solidly multi-platform aware
• remove most of the code from pkg/cmd/image/list.go and other places that do access low-level details directly
• come-up with serious test - we have close to 0 multi-platform tests for images

Additional context

No response