import: ignore non-successful HTTP codes for collecing image metadata

Previously we'd collect the data from redirects too, which wasn't
particularly terrible, since these typically don't carry the data we
were interested in, but it's still incorrect to do so.