Fix gdb.threads/access-mem-running-thread-exit.exp w/ native-extended-gdbserver

When testing gdb.threads/access-mem-running-thread-exit.exp with
--target_board=native-extended-gdbserver, we get:

  Running gdb.threads/access-mem-running-thread-exit.exp ...
  FAIL: gdb.threads/access-mem-running-thread-exit.exp: non-stop: second inferior: runto: run to main
  WARNING: Timed out waiting for EOF in server after monitor exit

		  === gdb Summary ===

  # of expected passes            3
  # of unexpected failures        1
  # of unsupported tests          1

The problem is that the testcase spawns a second inferior with
-no-connection, and then runto_main does "run", which fails like so:

 (gdb) run
 Don't know how to run.  Try "help target".
 (gdb) FAIL: gdb.threads/access-mem-running-thread-exit.exp: non-stop: second inferior: runto: run to main

That "run" above failed because native-extended-gdbserver forces "set
auto-connect-native-target off", to prevent testcases from mistakenly
running programs with the native target, which would exactly be the
case here.

Fix this by letting the second inferior share the first inferior's
connection everywhere except on targets that do reload on run (e.g.,
--target_board=native-gdbserver).

Change-Id: Ib57105a238cbc69c57220e71261219fa55d329ed