This test was originally starting with an empty set of checksums,
leading to the download of a checksum for every released and snapshot
version. This resulted in in sporadic test failures.
We now start with a known set of checksums and ensure that those that
are missing are downloaded. This involved some refactoring and
improvement in the way snapshot checksums are processed.
The checksum values for most wrapper versions are hard-coded into the
action. These known checksum values are first used for validation: only
if none of the known values work do we download checksums.
Previously, we blindly downloaded all of the checksum values in this
case: we now only download the checksums for versions that are not in
our "known" set.
Fixes#171
