It will allow smart cache detection «there is a file, so I can get it from cache»
and smart cache date usage «can we use that file? — no, it's too old»
mainly because it will avoid code duplication.
It will also enable better disconnection detection, as we can now handle the
«we got redirected to /enrol» in one place

For now, we keep download_file_async separate, because it avoids keeping the whole file in RAM.

Fixed remaining bugs to make selection viable, plus some other bugs (like segfaulting on close when all request aren't fulfilled yet because callback no longer points to some sensible adress)