remove pop_target

This patch fixes the target double-close problem (PR remote/15266),
and in the process removes pop_target entire (PR remote/15256).

The first issue is that pop_target calls target_close.  However, it
then calls unpush_target, which also calls target_close.  This means
targets must be able to be closed twice.  Not only is this strange,
but it also directly contradicts the contract of to_xclose targets.
(We currently have just a single such target, and it is never pushed;
but I plan to add more, and so this latent bug is triggered.)

The second issue is that it seems to me that calling pop_target is
often unsafe.  This is what cropped up in 15256, where the remote
target assumed that it could pop_target -- but there was another
target higher on the stack, leading to confusion.

But, it is always just as easy to call unpush_target as it is to call
pop_target; and it is also safer.  So, removing pop_target seemed like
an improvement.

Finally, this adds an assertion to target_close to ensure that no
currently-pushed target can be closed.

Built and regtested on x86-64 Fedora 18; both natively and using the
native-gdbserver board file.

	PR remote/15256, PR remote/15266:
	* bfd-target.c (target_bfd_reopen): Initialize to_magic.
	* monitor.c (monitor_detach): Use unpush_target.
	* remote-m32r-sdi.c (m32r_detach): Use unpush_target.
	* remote-mips.c (mips_detach): Use unpush_target.  Don't
	call mips_close.
	* remote-sim.c (gdbsim_detach): Use unpush_target.
	* target.c (pop_target): Remove.
	(pop_all_targets_above): Don't call target_close.
	(target_close): Assert that the target is unpushed.
	* target.h (pop_target): Don't declare.
	* tracepoint.c (tfile_open): Use unpush_target.

gdb/tracepoint.c

@@ -4366,8 +4366,8 @@ tfile_open (char *filename, int from_tty)
     }
   if (ex.reason < 0)
     {
-      /* Pop the partially set up target.  */
-      pop_target ();
+      /* Remove the partially set up target.  */
+      unpush_target (&tfile_ops);
       throw_exception (ex);
     }