Allow factorizing permutations into words in somewhat larger examples

[fingolfin]

Tweak limits for Minkwitz algorithm. The old bound were somewhat conservative on current systems. Esp. the total limit of 25000 was very low.

[hulpke]

Yes, this is a relic of olden times. Could even increase by a factor of 5 by now.

(There are other places where code sets memory usage limits. Do we want to have a general framework how much memory we consider appropriate for a routine to gobble?)

[fingolfin]

Yeah perhaps we should think about a more systematic way to track such "magic numbers" and limits so that we can scale them better over time...

In the meantime, I am surprised that this change causes a regression with a relatively small example in the manual, in lib/ghomfp.gd:

##  The following example stems from a real request to the ⪆ Forum.
##  In September 2000 a ⪆ user working with puzzles wanted to express the
##  permutation <C>(1,2)</C> as a word as short as possible in particular
##  generators of the symmetric group <M>S_{16}</M>.
##  <P/>
##  <Example><![CDATA[
##  gap> perms := [ (1,2,3,7,11,10,9,5), (2,3,4,8,12,11,10,6),
##  >   (5,6,7,11,15,14,13,9), (6,7,8,12,16,15,14,10) ];;
##  gap> puzzle := Group( perms );;Size( puzzle );
##  20922789888000
##  gap> hom:=EpimorphismFromFreeGroup(puzzle:names:=["a", "b", "c", "d"]);;
##  gap> word := PreImagesRepresentative( hom, (1,2) );
##  a^-1*c*b*c^-1*a*b^-1*a^-2*c^-1*a*b^-1*c*b
##  gap> Length( word );
##  13

Apparently with this change, instead of a word of length 13 it finds one of length 57?