Make succeeded and failed in ICB final + introduce onSuccess

[lorban]

Closes #11932

[lorban]

@gregw I went over all our ICB implementations that override succeeded and failed, and I came up with two conclusions:

• Overriding succeeded is often used to release a successfully written buffer before moving on to the next one. But that can easily be replaced by a new onSuccess protected method that can do the same job but in a safe way.
• Overriding failed is sometimes done to release a buffer, but that can be done in onCompleteFailure instead.
• There are two places where succeeded and failed are overridden for different reasons, and IMHO those are pure evil. See CustomTransportTest and H3's MessageFlusher to see what I mean: the 1st one calls iterate() from succeeded and does not call super, the 2nd one calls super.succeeded() from failed().

I'm now looking for a way to modify MessageFlusher so that it doesn't need to act this weird, but I'm building a firm belief that overriding succeeded and failed is never, ever a good idea.

[lorban]

I refactored MessageFlusher and make ICB's succeeded and failed methods final.

[gregw]

Nice big step in the right direction! But a few things....

[lorban]

@olamy do you have an idea why the integration tests are failing with a compilation error despite the fact that the code doesn't match what's reported?

[lorban]

@gregw I reverted the final modifiers in ICB, and I also fixed a small bug in the delayed onSuccess call and added a test.

I think this can now be merged.

[lorban]

@olamy thanks for your help fixing the build!