Tethering offload stats updates are eventually consistent

This patch removes the call to runWithScissors() in
OffloadController#getTetherStats() that was causing a deadlock when
NetworkStatsService would be polled for stats in certain threading
contexts.

Instead of trying to query the tethering offload HAL synchronously all
the time, this patch:
 - changes getTetherStats() to only call into the offload HAL when it
   detects that it is called on the same thread as the Tethering handler
   thread.
 - changes the map of interface to accumulated tethering forwarded stats
   to be concurrent.

This makes stats reading from getTetherStats() eventually consistent.
From the point of view of getTetherStats(), it preserves the guarantees
that tethering stats are monotonically increasing, and also guarantees
no tearing between rx bytes and tx bytes.

Bug: 29337859
Bug: 32163131
Bug: 64771555
Test: runtest frameworks-net
Change-Id: Ibcd351ad0225ef146b00a807833f76d2a886f6c1
(cherry picked from commit eb5e465e)