Are there any complete new techniques, algorithms, shaders broadly
used in next gen, already known on paper but not used in current gen
because of limitation of hardware or this is all same but more
complex,less generic or made for in-house use(unknown for public) to
provide better visual result for specified engine.
Mostly all about lighting models which required more capable shader engines and more GPU memory/bandwidth.
There are various other useful techniques like computer shader, tessellation, etc., but most of the "realistic wow" is all about lighting.
Improved texture fidelity also helps a lot. The 256/512M available on older console hardware was nowhere near enough compared to the PC.
What I expected to see but was not shown:
Fracturing Generic destruction of buildings and cars! Is this still to
compute heavy for large scale scenes?
Yes. These require complicated physics models and data-definitions. Some last-gen games like Force Unleashed did some "real" deformable physics but only in limited/specific circumstances, e.g. the doors you could "force bend" open.
Modeling the destruction of a building even semi-realistically using just physics for example would require physically modeling the structure and interior of the building as well as its outside "veneer." We're talking a huge increase in content creation time and cost. Destruction can be faked a bit more easily by creating pre-broken versions of assets and then using physics to just move about the pieces, but this both looks wrong for most non-simple objects (it's done for rocks and trees and the like in many games, but not buildings or cars or complex objects) and is still a huge time/money sink compared to just not having destructible terrain.
Games are all smoke and mirrors. The world simulation you see is faked to every extent possible. There's a limit to how "realistic" games are going to get any time soon.