Instead of reading the central dir and then the local dir (and trying to
handle all the cases, data descriptors, data descriptors with zip64, etc)
I just read the central dir, hoping that everything is here. (The only
exception is the file name : for some reasons, some zip have the real
name : with / instead of \ in the local dir)
This fixes an issue with nested zip files and data descriptors. When the
size was 0 in the local dir, we would seek the next occurence of the
data descriptor signature. With zip containing data descriptors, this
would lead to a bug.
This should also help the issue #30 : the new code avoids the overhead of
looking for data descriptors.
Other change : I explained the limitations of the zip64 extensions in