Turns out PBD is not very suitable for simulating absolutely rigid bodies (?).
I quickly had spheres jumping around on a plane, but couldn't get rid of jitter (becomes visible if increase gravity or decrease the timestep).
Maybe, that's why they solve contacts on velocity level (like in most impulse-based engines).
As for constraints equations, these lectures are extremely easy-to-understand:
[2014, 'so this is my *ss', lol, sorry]: