[Patch, Fortran, committed] Improve diagnostic for I/O on alloc_comp/pointer_comp

[Tobias Burnus]

Before it was not obvious why the compiler rejected derived types with 
allocatable or pointer components; cf. recent discussion in 
comp.lang.fortran (and at j3). This patch mentions defined input/output 
to make clear how to solve it. (Or not, given that defined I/O is not 
yet implement in gfortran.)

Additionally, I have updated a comment to correctly point to the renamed 
function.

Committed as Rev. 176386.

Tobias