Avoid CI failures because of download failures

[exaexa]

Models are downloaded from the corresponding repositories each time, which may be a problem for repositories not backed by CDNs (as with EBI) because random interruptions/failures there cause the failures on our side.

We can cache downloaded data with actions/cache as done now with Julia artifacts to avoid much of this (thanks @giordano for suggestion)

Also might need to revamp the logic of downloading and checking hashes, to prevent failures from caching bad data.

For reference, 1 "stable" and 2 "nightly" runs seem to have failed here because of download problems: https://github.com/LCSB-BioCore/SBML.jl/actions/runs/2479749266