Prevent asynchronous calls from hanging when service crashes

This CL introduces hidl death recipients to prevent asynchronous calls
from hanging when a service dies. The general process is as follows:
(1) Register HIDL death recipient ("DeathHandler") when an interface
    object is constructed.
(2) Register I*Callback object with DeathHandler immediately before an
    asynchronous execution.
(3) Immediately wait for the callback object to be populated with either
    (a) the results of a valid asynchronous execution
    (b) the error code provided by the DeathHandler if the service has
        died
(4) Unregister I*Callback object with DeathHandler after the callback
    object contains a result.
(5) Repeat 2-4 as many times as necessary.
(6) Unregister HIDL death recipient ("DeathHandler") when the HIDL
    interface object is being destroyed.

As an extra convenience and protection, #4 (callback unregistration)
happens automatically when the current scope is exiting via an
RAII-manager "ScopeGuard" object.

Fixes: 118624080
Test: ran "NeuralNetworksTest_static  --gtest_filter=GeneratedTests.add",
killed the sample-all driver, and confirmed (1) that the runtime was not
blocked and (2) that the appropriate log message was recorded. NOTE: this was
facilitated by adding a 10 second sleep in the sample driver for the
asynchronous preparation and asynchronous execution, enabling the service
to be manually killed via "adb shell kill -9 <pid>".

Change-Id: Ie90de41eadfe0d8a1e71f5079962f27ae97852cb