Implement fastdg for systems in the "new" style

When implementing the new component-wise style in !164 (merged), we got sumfactorziation for systems quite cheap, but I forgot to implement fastdg for those cases as well (which requires the usage of offsets into the data array). there is good chances we already have a working or quasi-working implementation in !153 (closed) (at least we already patched loopy to do it!)