Prevent double teardown of service connections

Asynchronicities in activity teardown -> service connection teardown
introduced a race in which the teardown could race with new service
bindings to "the same" service instance, and then wind up attempting to
shut down a new, valid instance inappropriately.  Fixed by making sure
to clear the "what needs to be torn down" bookkeeping as part of the
act of doing that teardown, removing the possibility for stale state.

Fixes: 131029480
Test: manual
Test: atest CtsAppTestCases
Change-Id: I33a63f524d147ff6ec97dd3efb0127dcace8bf3c