ARM long branch stubs: cleanup

[Paul Brook]

> > > With the attached small patch, I propose to remove an error-prone
> > > switch statement to get the correct stub template and size in
> > > arm_size_one_stub(). It is replaced by an (error-prone :-) array
> > > defined next to the stub themselves.
> > >
> > > Maybe it would save some maintenance issues.
> >
> > If you rework the preprocessor macros a bit you can generate both the
> > enum and the array from the same list.
>
> But then you get backslashes all over the place - is it an
> improvement?

Hmm, maybe? It doesn't look as pretty, but means you don't have to maintain 
duplicate lists.

Paul