That's the benefit of hardware failsafe. Something could break to leave an LED on,
but the watchdog timer is still the cheapest and most effective on the software side.
If your display routine fails to cycle, the watchdog should be on an independent timer,
and fire even if, for example the chip's oscillator shorted.

If you have rows of 12 LEDs parallel, that means they can't be independently controlled
for graphics, etc. You are just cycling whole rows of LEDs?

The current can still be measured at 100% duty cycle, and as you lower the duty cycle,
the average current should lower approximately with the duty cycle.
Transistors have cut off slopes, so it's not extremely precise. Someone else may have to chime in about that.
However, just because you use say 50% duty cycle, the fact remains that the row of LEDs and hardware to drive
them are using the same current for half of the time, and no current for the other half of the time.
That's if you negate a little error for the actual switching on and off, but I don't know exactly what the cost is there.