The signal theoretic framework depends on bandwidth limitations that often don't hold for signals we want to produce. Sharp lines and boundaries are reasonable things to want.
I'd say the sampling kernels aren't fully matters of taste, but matters of the display or sampling technology instead. Pixels on CRTs are indeed not little squares, but pixels on anything else pretty much are -- though they can be non-contiguous squares, with different patterns for different colors, even, which eliminates any simplicity that the little-square picture intuitively captures.
Fortunately, all this starts to matter less and less as pixel size and spacing gets smaller and smaller.
I'd say the sampling kernels aren't fully matters of taste, but matters of the display or sampling technology instead. Pixels on CRTs are indeed not little squares, but pixels on anything else pretty much are -- though they can be non-contiguous squares, with different patterns for different colors, even, which eliminates any simplicity that the little-square picture intuitively captures.
Fortunately, all this starts to matter less and less as pixel size and spacing gets smaller and smaller.