The "individually exposed sensel" sort of mirrors the "individually backlit LCD pixel" display. It seems to me that you just move the problem from one place to another (potentially a good or bad thing).
In the case of LCD tvs, having one backlight per pixel seems to not be cost-effective; the LCD panel would no longer be needed if you go all the way to the subpixel, and what you have is effectively a LED-tv, not a LCD-tv. What is cost-effective, though (or at least quality-boosting) is to have individual LED backlights operating on groups of LCD pixels. Since many/most images are kind of smooth/lowpass (with low-contrast, high-frequency local modulation vs high-contrast, low-frequency large-scale modulation, or at least our vision seems limited to such an interpretation), it makes sense to approach the high-resolution/high-contrast ideal with this compromise where you get either high-resolution or high-contrast (but not both at the same time).
The GND-filter does something similar: part of the scene is made darker, thus, the DR is compressed before entering the sensor. This comes at the cost of reduced total exposure (unless you compensate by increasing time/aperture). One could expand this concept by inserting a sort of LCD panel in front of the sensor, and control it using the liveview output (attenuate light in areas that are brighter).
Without specific knowledge of sensor technology: if individual readout/reset of sensels is "hard", might it be possible to make them auto-reset? Once the well has reached its capacity (no sooner or later), its charge is dumped (into "the garbage"), and it starts accumulating once again. When the integration period is over, you read every sensel like today, but their values would be different: sensels that previously would have been saturated, are now "wrapped around" like a phase value. By using clever 2-d unwrapping, one might have an estimate of the true sensel values (assuming that most scenes are smooth, just like highlight recovery).
-h