Fix up some issues I found when porting PPC to the new common gdb stub code:

Fix up some issues I found when porting PPC to the new common gdb stub code:
- cosmetic changes in the messages printed
- 'flags' must always be unsigned long.
- explicitly calling initialize_gdb() is not difficult. For x86 and ia64 I
placed this call immediately before do_initcalls(), since that's where it's
being called from now so we know it's safe. Architecture people can move it
earlier as appropriate.
- I don't understand all these ASSERT(!local_irq_is_enabled()) statements,
sometimes bracketing a single call like receive_command(). How exactly would
receive_command() manage to re-enable irqs? Also, a failing ASSERT would just
call into the stub again anways...
- initialize_gdb() was overcomplicated. serial_parse_handle() already handles
the parsing for us, and there's no need to panic there.

Signed-off-by: Hollis Blanchard <hollisb@us.ibm.com>