libimage: do not fail disk usage when an image is removed mid-walk

[vtushar06]

podman system df flakes in parallel CI runs when another test removes an image while df is walking the list:

# $ podman system df --format {{"\n"}}
# Error: Image f995a32fc412 exists in local storage but may be corrupted (remove the image to resolve the issue): reading image "f995a32fc412...": locating image with ID "f995a32fc412...": image not known
# [ rc=125 ]

(from https://github.com/podman-container-tools/podman/actions/runs/29565253194/job/87336101705, and again on 07-21)

DiskUsage lists the images once, then diskUsageForImage calls isCorrupted per image, so anything removed between the two steps kills the whole command. Same shape as the volume flavor that 246724fc6 fixed on the podman side, just on the image path.

Two small changes: isCorrupted wraps with %w instead of %v so the error keeps its type, and DiskUsage skips an image only on ErrImageUnknown. A genuinely corrupted image still fails - TestCorruptedLayers builds the missing-layer case and it passes unchanged, which I checked matters here because that error is ErrLayerUnknown, not ErrImageUnknown.

The new test fails on the old code (require.ErrorIs cannot see image not known through the %v wrap) and passes with the fix. Ran both tests on linux in a container.

[vtushar06]

@giuseppe I think this will be ready, just waiting for workflow to pass out

[Luap99]

I think the code change itself is fine, but since this is a race we cannot really that this like that. IMO we could drop the test as loing as all other pass that should be fine.

[vtushar06]

@Luap99 yeahh right as the delete happens before DiskUsage lists the images, so it never walks that one and the assertion passes with or without the change. dropped the test, the two code commits are unchanged. let me know if everything is fine

[Luap99]

LGTM

[Luap99]

@giuseppe PTAL again

[giuseppe]

LGTM