adjust generation of component addresses

[Richard Kenner]

    > But with your change, the N_ARRAY_REF returns a MEM whose address is
    > a PLUS of the stack address and an index.

    I presume from this comment I should be trying a target that
    has reg+reg addressing, like x86?

Well I first saw it on x86, but thought it appeared elsewhere. Maybe not.

    Err, which calls?  In the Ada test case?  I.e.

       Print_Values (B_Value =>Get_B(J), A_Value =>Get_A(J));

    or something else?

That one.

    What wouldn't be fixed by adding yet another call to
    update_temp_slot_address in offset_address?

You're right: that does work.  I don't know why I didn't think of it.
I'll check that in and we'll see what's still broken during tonight's
runs.  Thanks.