I gave up on I2C reads (so, if someone gets them working, I'd still love to hear).
I tried UART mode, which eventually worked fine. The problem there is that the 78K baud boot messages (which, apparently, cannot be turned off, based on what I found in other threads) confuse the display. The easiest workaround that works 99% of the time is: Serial.setDebugOutput(0), a small delay, then printing a couple of newlines, before starting to issue commands. However, Digole support told me that the reset signal, although not broken out, is easily accessible via the unpopulated reset switch pads. So, the best solution is to solder a wire to one of these pads (on my board, these were the bottom two -- check with a multimeter which ones are not connected to ground), and drive it low after bootup to reset the display. Works every time.
As for the picture, after some extended Googling, it appears to be a moderate case of non-reflow: http://www.bobwillis.co.uk/ppm/defects/ ... Nr48tf.asp An additional hint was that some of the sdcard slot's holding pads are popping off (insufficient wetting too?). It does not seem to be affecting the function of the board, though.