Arvados

When the cached discovery document contains a broken URL, the arv cli continues to try to connect to it anyway and does not remove the cached document. Thus, in order to restore function of the arv cli after a bad discovery document is cached one has to manually remove it (i.e. `rm -rf ~/.cache/arvados/discovery*`).

I would expect the cache to be invalidated if one of the given URL results in a connection error.