Empty image cache metadata

[chanseokoh]

#534 (comment)

It's worth an investigation. Even if this happens only when there is a problem interacting with a registry, we should be able to have a correct cache metadata locally.

[coollog]

This is due to the build failing at the manifest push step, whereas cache metadata is currently written after build steps succeed. To resolve this, we would need to either

1. Change cache metadata to be independent for each layer so that each can be written independently and during their respective build steps, or
2. Catch any failures during the build steps and write the cache metadata upon failure as well

I prefer approach 1) since it is more robust and less prone to possible cache corruption.

[Hi-Fi]

Reason I tried to check the cache was to see what kind of Manifest Jib tries to send (and compare it to one that's obtained from Artifactory/Red Hat Container Registry).

[chanseokoh]

@Hi-Fi FYI, the Jib cache metadata is just an Jib-internal data structure and has nothing to do with files actually used for the Docker or OCI image format.

[coollog]

This will be fixed by the new cache mechanism #878