gdb/arc: fix print_one_insn selftest

I noticed that the ARC print_one_insn selftest was failing.  The
problem is that in print_one_insn_test the arc case falls through into
the special case that handles nios2, score, and riscv.

The special case for these targets hard codes the breakpoint kind to
'4'.  This is find for bare metal arc (see arc-tdep.c,
arc_sw_breakpoint_from_kind), however, for arc/linux only breakpoint
kind '2' is supported (see arc-linux-tdep.c,
arc_linux_sw_breakpoint_from_kind).

So the code in print_one_insn_test as it is currently written passed
in an invalid breakpoint kind, this leads to GDB trying to disassemble
unexpected memory.

The fix is to reorder the code in print_one_insn_test so that the arc
case falls through into the default case.  In the default we no longer
hard code the breakpoint kind, and everything should be good.

Additionally, given the arc code only expects specific breakpoint kind
values, I thought it would be nice to add some gdb_assert to validate
things.  This assert would have triggered in this case and made it
easier to find the error.

After this commit, configure GDB with --enable-targets=all, then run
gdb.gdb/unittest.exp, there should no longer be any failures.

gdb/ChangeLog:

	* arc-linux-tdep.c (arc_linux_sw_breakpoint_from_kind): Add an
	assert.
	* arc-tdep.c (arc_breakpoint_kind_from_pc): Likewise.
	* disasm-selftests.c (print_one_insn_test): Fall throough from ARC
	case to the default.

gdb/arc-linux-tdep.c

@@ -149,6 +149,7 @@ static const gdb_byte *
 arc_linux_sw_breakpoint_from_kind (struct gdbarch *gdbarch,
 				   int kind, int *size)
 {
+  gdb_assert (kind == trap_size);
   *size = kind;
   return ((gdbarch_byte_order (gdbarch) == BFD_ENDIAN_BIG)
 	  ? arc_linux_trap_s_be