D2 is just reverse polarity protection on the negative supply, so the other 'GND' is normally -7V, not GND. The diode was setup that way to allow one to place a shunt between pin 5 & 6 of the 6-pin connector to use the device with a single supply if you don't need common-mode voltages near 0V. In which case AGND becomes the GND you were mentioning.
The lack of precision regulators for the offset trimmer and buffers are due to the original use-case of this. Basically when measuring current variations for attacks w/ the ChipWhisperer project, you don't care about small changes in gain, non-linearity, or offset. The device this plugs into is always AC-coupled, and data is frequently normalized anyway.
So these aspects were omitted to make the design easier for someone to build dead-bug style (as my prototype was). Since the final boards had the same use-case I never bothered to change anything (and the design worked so didn't want to change it any more).
As an interested side-note the
eval board for this device uses the same approach (just an unbuffered resistor), even though the datasheet had called out that the source impedance of the offset voltage should be low.
Perhaps there would be a market for a slightly better version for people wanting to use this as a generic scope probe....