whack-a-mole: no need to set_fs(USER_DS) in {start,flush}_thread()

flush_old_exec() has already done that.  Back on 2011 a bunch of
instances like that had been kicked out, but that hadn't taken
care of then-out-of-tree architectures, obviously, and they served
as reinfection vector...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

arch/arc/kernel/process.c

@@ -155,8 +155,6 @@ int copy_thread(unsigned long clone_flags,
  */
 void start_thread(struct pt_regs * regs, unsigned long pc, unsigned long usp)
 {
-	set_fs(USER_DS); /* user space */
-
 	regs->sp = usp;
 	regs->ret = pc;