There's softblock and hardblock, rfkill shouldn't be able to unblock hardblocks. "unblock all" just attempts to soft unblock all devices. Maybe it could also operate some special software operable hardblock hardware but that doesn't count.

Also/IIRC, the old webcam issue was some cameras had ~1s delay between first image to LED on, and zero for last frame to LED off, so initialization or configuration change commands could be mashed to allow LED to light in user unhelpful ways: e.g. only lit for imperceptible length of time or intermittently lit as shown. Apple and few other vendors modified this behavior as well as added a minimum LED on duration while after this went on media. It's still "controlled by software", sure, not like tapped off of CMOS imager power supply, but not like implemented in /lib/modules/webcam_comforting_led.so.1, either.

I was able to use the rfkill command on a Dell and a Thinkpad with the hardware switch turned off. However I don't know how that was implemented internally, maybe the switch was not connected directly to the mini-pcie slot. I don't have these laptops anymore so I can't check it.

I made the video on the webcams, these are quite modern ones. Nearly all of them have special UVC commands that allow reading and writing the internal CPUs memory (a lot are 8051 based). You can find the register that controls the led GPIO and control it freely. Only on one of them I did not find a way to modify the led behavior without modifying the firmware to add an extra command.