Aborted sgr clone --download-all causes future clones to redownload all objects

[mildbyte]

Repro:

• try cloning a big repo with --download-all (e.g. splitgraph/socrata) -- fails because of no space in cache
• try cloning just the :latest (sgr clone splitgraph/socrata:latest --download-all), fails with weird object size accounting:

Gathering remote metadata...
No image/object metadata to pull.
Need to download 1 object (34.53 GiB), cache occupancy: 1.99 MiB/10.00 GiB
error: splitgraph.exceptions.ObjectCacheError: Not enough space in the cache to download the required objects!

Deleting the offending repo locally + sgr cleanup fixes it