Hell, my solar panel (which reacts slower than a photodiode) has its response to green/blue phosphors die in 1-2 480p scanlines (which last half as long as 240p scanlines)! Though if hooked up to a high series resistance, it takes a looot longer for the charge to dissipate.
I suspect the difference in switch-off times between a Zapper and Light Phaser is related to the support circuitry (https://hackaday.io/project/9782-nes-zapper-video-synth-ther..., capacitors? transistors?!) translating the photodiode's signal to the controller pin. Series capacitors will block continuous light, and higher resistances will make a short signal last longer before dissipating.
> This would take a lot of frames to check every single letter if this worked the same way as the Zapper.
You could binary-search to find which letter was shot, though this would obviously fail if the user is moving the light gun between letters. It would probably be more reliable if the lit-up region wasn't just a small box but the entire grid cell.
> But look at what we’ve done here: the Master System is literally reading in-line with the electron beam. This means that for this to work, you really do need an electron beam that moves in sync with the VDP. An LCD just won’t give you that moving dot, no matter how much delay you add.
What's the latency of CRT composite chroma decoding? (EDIT: http://www.hawestv.com/mtv_color/delayline.htm says luma is delayed by 1.3 us to match chroma. Compare to the total scanline duration of 63.7 us, or visible duration of around 51 us.)
How many pixels does that shift the light gun? (1.3 / 51) * 256 px = around 6.5 pixels. It's possible that Master System games compensate for this delay using an empirically tuned horizontal offset.