I've worked around it by sending that very first command mode byte twice in a row. This reliably makes everything work - at first power up, or during any hard or soft reset. Other than initially on first power-up, I never see any corruption on the display, so I'm fairly confident I've got all the timing right (believe me, it wasn't at first) along with the clock polarity etc. (and I know it works without this hack on other MCUs - so other than a few Teensy differences in how the SPI hardware works, the routines are basically correct).