Exporter should invalidate the cache if it fails to write it

[tomkennedy513]

Summary

Currently, when the exporter fails to write the cache after successfully exporting the image, it emits a warning without failure. This is acceptable behavior in my opinion, but when the caching fails, the exporter should at least invalidate the cache so that we don't have a corrupted cache causing issues in subsequent builds.

---

Proposal

The exporter should invalidate the cache whenever it fails to successfully write it, or if the cache it did write is somehow not valid.

Context

We are seeing restore failures in kpack due to a corrupt cache after the exporter only writes out a partial cache before failing.

[jabrown85]

In the 10/17 meeting

We talked about adjusting restorer to check and fail if the cache doesn't match the metadata expected. Adding in this verification will not require a spec change unless we find performance or other issues during development of the fix.