[patch] Fix gdbarch update logic

[Andrew Cagney]

Hello,

The attatched patch fixes the logic used to select a new architecture. 
It should fix two bugs:

Given the situtation:

	gdb executable core

where bfd couldn't recognize the core, multi-arch was trying to switch 
the architecture to ``unknown'' instead of just retaining the 
architecture identified in ``executable''.

Given a user entering a command like:

	(gdb) set architecture foo
   or	(gdb) set byte-order big

multi-arch, was for some situtations, ignoring this and instead using 
the architecture supplied by the INFO struct.


This patch changes the logic selecting an architecture so it uses the 
selection sequence:

	Hard wired by (gdb) set ....

	Privided by the INFO struct

	The previous architecture

Andrew