I've been looking at a walkthrough of the Cyberpunk 2077 world.

No gameplay, just a long look at the models, lighting, and rendering. Models are OK, lighting is very good.

They do a nice job when looking into a dark place from bright sunlight, or out of a dark place into bright sunlight. Watch starting at 17:00. You can see a sort of “auto-iris” effect going on. How is that done? Are they using high dynamic range rendering followed by post-processing, or what?

Looks like conventional auto exposure to me. I never worked on this, but it's about making a histogram of brightness from the frame, then calculating a mapping to bring the image into the small range our current displays have. Temporal exponential average can be used to avoid harsh changes, so the transition is smooth like it is for human vision.
If we had true HDR displays, we would not need to do this. Some numbers i remember: A white wall outdoors, lit from direct sun can have 10000 nits, and current HDR displays may have 500, conventional displays 200. (I may remember this wrong, but just to give some impression.)
Ofc. we need to render at high precision, which is now standard but would have been hard with 8bit framebuffers.