fix a mutability inconsistency of ExtRepOfObj values of objects in IsLetterAssocWordRep

[ThomasBreuer]

Up to now, we get

gap> f:= FreeGroup(IsLetterWordsFamily,4);;
gap> IsMutable(ExtRepOfObj(f.1));
true
gap> IsMutable(ExtRepOfObj(f.1));
false

The point is that an immutable version of the result gets cached after the first call, and this cached version is used in the second call.
With the proposed change, also the first result is already immutable.

[fingolfin]

This promptly broke the test suites in semigroups (ping @james-d-mitchell) and groupoids (ping @cdwensley). At least that is how I interpret the test failures for groupoids and semigroups.

[james-d-mitchell]

@fingolfin interesting, this should be easy enough to fix, will try to do it today.

[cdwensley]

@fingolfin: groupoids fixed - version 1.76 released