Download only the package header, not complete RPMs.
    
    Currently, the rpmdpelint downloads the complete RPMs just to load
    their headers to check for file conflicts. This is waste of time
    and resouces.
    
    This commit changes this to download only RPM headers.
    
    There is no function provided by the Python `rpm` module which would
    return the size of RPM header. The code therefore tries to download
    first N bytes of the RPM file and checks if the header is complete or
    not using the `hdrFromFdno` RPM funtion.
    
    As the header size can be very different from package to package, it
    tries to download first 100KB and if header is not complete, it
    fallbacks to 1MB and 5MB. If that is not enough, the final fallback
    downloads whole RPM file.
    
    This strategy still wastes some bandwidth, because we are downloading
    first N bytes repeatedly, but because header of typical RPM fits
    into first 100KB usually and because the RPM data is much bigger than
    what we download repeatedly, it saves lot of time and bandwidth overall.
    
    Checksums cannot be checked by this method, because checksums work
    only when complete RPM file is downloaded.
    
    Signed-off-by: Jan Kaluza <jkaluza@redhat.com>