Replace deprecated Throwables.propagate with throwIfUnchecked and direct exception throwing in SubscriberRegistry, MoreExecutors

[panic08]

The suggestion here is to change the use of the deprecated Throwables.propagate() in the SubscriberRegistry.java, MoreExecutors.java classes to the alternative that is recommended in the documentation for this method:

Throwables.throwIfUnchecked(e);
throw new RuntimeException(e.getCause());

[cpovirk]

I wanted to make sure that I came back and said something here because this PR prompted a bunch of activity :)

• The PR had a subtle error that we've seen before in each file: Instead of throwIfUnchecked(e.getCause()), it had throwIfUnchecked(e).
  • That actually causes a build error in one case (the InvocationTargetException case) because our build runs https://errorprone.info/bugpattern/ThrowIfUncheckedKnownChecked. (Sadly, running our build is still slightly more complex than it should be, but ./mvn clean install should normally do the trick.)
  • It's not a build error in the other case (the UncheckedExecutionException case) because Error Prone doesn't have a checker for a known unchecked exception. (There would probably have been a test error, at least.) I'm proposing adding such a checker in https://github.com/google/error-prone/pull/4625.
  • I think it was fear of that kind of problem that had led us to employ some awkward code elsewhere, so I tried out a new pattern for avoiding such problems in https://github.com/google/guava/pull/7442.
• The existing behavior in each class was questionable:
  • SubscriberRegistry was rethrowing an exception that could (in rare cases) come from another thread. This can be confusing, since the stack trace in that thread doesn't necessarily match the stack trace in the thread that eventually catches the exception. I ended up changing that in https://github.com/google/guava/pull/7449, and I just sent up a followup in https://github.com/google/guava/pull/7453.
  • MoreExecutors was more reasonable, but I've come to view "throw new RuntimeException(e)" as a mild code smell: While it's very rarely actually important to throw a specific subclass of RuntimeException instead, "throw new RuntimeException(e)" has historically sometimes indicated that the author just didn't want to think about what might have caused the exception or what should actually be done about it. We ended up adding a comment and using UndeclaredThrowableException, and then I made similar changes in some other files in https://github.com/google/guava/pull/7451, plus https://github.com/google/truth/pull/1353 in lieu of a similar change elsewhere (which then tipped me off to search for some other Java 7 / Android updates in https://github.com/google/guava/pull/7446 / https://github.com/google/truth/pull/1352).
• Given how much I ended up jumping from place to place, I figured that I'd follow up on some other existing warnings, for which you can see further changes in https://github.com/google/guava/commits/master/.