NetworkReadStream actually saves whole response in the memory now.
There is a pause mechanism in libcurl, but if libcurl is requesting
something compressed, it would have to uncompress data as it goes even
if we paused the request. Even though our own stream won't be notified
about this data when when "pause" the request, libcurl's own buffer
wound be expanding.
CurlRequest uses own multi_handle, in which it creates an easy_handle to
make a request.
Every time `handle()` is called it checks whether request is complete
and, if it is, stops.
