Improved complex typed supervision definitions

[johanandren]

Supervising a typed actor with many different strategies for different exceptions currently creates a "pyramid of doom":

val myActor = system.spawn(
      Behaviors
        .supervise(
          Behaviors
            .supervise(
              Behaviors
                .supervise(Behaviors
                  .supervise(MyActor())
                  .onFailure[ExceptionA](SupervisorStrategy.stop.withLogLevel(Level.ERROR)))
                .onFailure[ExceptionB](SupervisorStrategy.stop.withLogLevel(Level.WARN)))
            .onFailure[ExceptionC](myRestartStrategy.withLogLevel(Level.WARN)))
        .onFailure[ExceptionD](myRestartStrategy.withLogLevel(Level.ERROR)),
      "my-actor")

It would be nice with an API that flattens such a set of supervisions to a shorter one that is easier to parse, perhaps something like:

val myActor = system.spawn(
      Behaviors
        .supervise(MyActor())
          .onFailure[ExceptionA](SupervisorStrategy.stop.withLogLevel(Level.ERROR))
          .onFailure[ExceptionB](SupervisorStrategy.stop.withLogLevel(Level.WARN))
          .onFailure[ExceptionC](myRestartStrategy.withLogLevel(Level.WARN))
          .onFailure[ExceptionD](myRestartStrategy.withLogLevel(Level.ERROR))),
      "my-actor")

[patriknw]

I agree. This came up previously in a discussion with @manuelbernhardt and then the suggestion was also the flat syntax.