While subEntity(i,c,j,cc) is surely not allowed for
cc<c because there's no viable j, there's no reason
to not allow this case for size(i,c,cc). Actually the
implementation works nicely in this case and returns
the correct number size zero. Any grid-specific
ReferenceElement implementation should simply return
zero in this case.

Allowing this case gives the possibility to seamlessly test
for(int j=0; j < re.size(i,c,cc); ++j)
for all 0<=c<=dim and 0<=cc<=dim.