PRs 18769 and 29786

[Paul Thomas]

Since Brooks took an interest and offered to develop these PRs further, 
I have reassigned them to him.

Sorry, Brooks :-)

The fundamental part of both problems is to lay out the gfc_expr's for 
the initializers/arrays into a signed char array and then to 
reconstitute it into a field in the union or another gfc_expr with the 
appropriate typespec and arrayspec.  This is more or less done for 
PR29786 (modulo testing with different endian-nesses) and, once 
accomplished, the way looks clear to do PR18769

I will give Brooks every help/advice/assistance that I can but I have to 
bail out now.

Paul

PS Brooks, if this is too much, reassign them to me; we should flag that 
a fix is in progress.