2.x: Flowable as a Publisher to be fully RS compliant

[akarnokd]

This PR performs the changes suggested in #5110.

• Introduce FlowableSubscriber with extra textual specification on its relaxed nature
• Flowable.subscribe(Subscriber) checks for FlowableSubscriber and if not found, it wraps the incoming RS Subscriber into a StrictSubscriber that follows the RS spec to the letter at any cost.
• Introduce Flowable.subscribe(FlowableSubscribe) that most internal operators will use
• Change AbstractFlowableWithUpstream to accept Flowable as a source, update operators
• Some operators were useful with raw Publisher input, these were duplicated on their outer containing type but use the same internal FlowableSubscriber
• Removed "cheat" from the TCK tests, adjusted timeout on delay
• Replaced most implements Subscriber with implements FlowableSubscriber
• Replaced most new Subscriber with new FlowableSubscriber in tests, the rest is required for testing the strictness itself.
• strict() is now an identity operator with suggested scheduled removal.

Performance impact estimation

• Most primary use of a Flowable should go through subscribe(FlowableSubscribe) and thus no overhead change.
• Where the API mandated Publisher as input, providing a Flowable will have an instanceof check at subscription time and routed to `subscribe(FlowableSubscriber) if the consumer is part of RxJava 2 itself.

[JakeWharton]

Looks good!

[akarnokd]

Okay, let's go from here. In the unlikely case the spec becomes more forgiving, the undo is just to remove cast/wrapping from Flowable.subscribe and everything works the same as FlowableSubscriber is an interface.