[patch, fortran] Implement constant-folding for TRANSFER intrinsic.

[Paul Richard Thomas]

Brooks,

As you might have seen, I committed the patch in our collective name
and cleared the PRs.  It is a tribute to the excellence of you
contribution that, thus far,..... not a dicky bird.  At this stage,
after the gfc_conv_array_transfer patch there were already two PRs in
the in-tray.

I am pretty sure that my fix for derived types is a "kludge too far"
but it gives us a bit of time.  I am pretty sure that I can break it
and, when I have one or two things out of the way, I will try.  That
said, I know how to fix it too:)

What I am concerned about here is derived type association.  I am
rather sure that I can devise a mixture of initialization expressions
and use association that will trigger an ICE.  If I am right, this wil
come about because the derived type that is built in
gfc_simplify_transfer, simply comes at the wrong time.  The fix will
either be to NULL the ts.derived->backend_decl's or to bump them into
the golbal context and to ensure that other, like derived types pick
up the backend_decl.  I am warning you of this just in case I am in
one of my AWOL phases when it comes up:)  I will try to anticipate it,
though.

Anyway, good work - what else can we do together?

Cheers

Paul