Fix crash due to race on close()

It's possible for close to be called during local cleanup concurrently
with the remote call.

A recent change (ag/15993920) added a missing Binder.unlinkToDeath
call to close(). This now causes a crash now if the registration
was removed at just the right time in another thread. This synchronizes
access to close() to avoid this.

Bug: 237406501
Test: atest ScrollCaptureConnectionTest
Change-Id: I0126bfac1efdece2e4eff144a44f29a963553b74