  - Tensor und Subfield tests fail because SL(4,5) does not find an SL2
  - Use RECOG.RandomSubproduct everywhere
  - FindSL2 does not work for q=5, ==> SL(6,5) does not work (FindSL2)
    reason: we power up to order 4, square of y is -1 thus y and z=y^x
            mod -1 is dihedral rather than SL(2,5)
  - Look at Product Replacers again (--> orb): no need for accus during mixing
  - genss: Make call to StabilizerChain which is guaranteed to be quick
    possible
  - Think whether the changing of gensN after creation might hurt other
    places than LowerLeftPGroup.
  - SLPforElement methods without evaluation --> like in SiftAsWord ???
  - Convention: SLPforElement guarantees that a returned slp does the job
    ==> \in is easier, does not have to evaluate the slp
    ==> in StabChain code we divide up labels
    ==> in Hom code we divide up preimage of the image, go to kernel
    ==> doable everywhere (in fact done everywhere)
