Rather than checking the json file to make sure posts have been
successfully downloaded, data is stored in a temporary file which
is renamed when downloading has finished, as suggested in #257.
I've come across several implausible values for `device_timestamp` (such as 182428140, which would be in october 1975, assuming millisecond). I guess it is due to improperly configured phones, or maybe some third party software that's mangeling the EXIF data on images before posting. Anyway. Since by the nature of stories, the `taken_at` timestamp (presumably when the instagram servers received the post) ought to be approximately when an image was actually taken. So there's no real value trying to use the timestamp provided by the photo-taking-device.
The default Instaloader headers aren't passed with a simple `requests.head()`, so leakage of user agents such as `"python-requests/2.18.1"` will occur.
The original method of substituting 2048x2048 for whatever resolution was given seemed somewhat convoluted. This accomplishes the same thing, except raising an exception if the given url is not on the right domain.
