Everything I just wrote was true, but perhaps less helpful than it could have been...
1) For random reasons the WiFi on my test 4B was not enabled, and as reported the keyboard was working fine. Enabling WiFi killed the keyboard at some point during or after the scan - connecting wasn't necessary to break it.
2) Although the Bluetooth and WiFi halves of the CYW43455 are largely separate, there is a channel between the two for co-existence data. I suspect it is this mechanism which is failing, because rolling back the WiFi firmware solves the problem (which is odd, because this firmware was billed as improving coexistence, and I've not seen the problem in testing). The three older WiFi firmware files can be downloaded from here: https://github.com/RPi-Distro/firmware- ... c5f9c/brcm
- they're called brcmfmac43455-sdio.* . Install them the same way as you do the BT firmware - "sudo cp brcmfmac43455-sdio.* /lib/firmware/brcm" and reboot.
3) For some reason, despite seeing the hard error earlier, BT and WiFi do sometimes coexist on the latest firmware, which might explain how it wasn't spotted earlier. However, downgrading the WiFi firmware and leaving the Bluetooth firmware up-to-date is consistently reliable for me.