Tighten the checks around a detector being invalidated

Don't call back for a detector being marked invalid because
that happens when someone else obtains a detector or VIS shuts down,
in either case we don't want a loop where two entities keep creating new detectors
and being invalidated.

Don't call back on an invalid detector for availability change/detected/started and stopped
only propagate errors.

This helps us with cases where a callback for the previous VIS may get called and then crash because it
tries to make calls without being the current VIS.

In the new scheme of things, if the VIS changes, or the current VIS obtains a new AlwaysOnHotwordDetector,
the previous one is shutdown and internally marked as invalid and all calls to it fail with an IllegalStateException.

Bug: 16629417
Change-Id: I74417bf76ba80916ebc21b042c18b3467857733e