Yeah, not so much that it is impossible, just that it is limiting.
I've settled on a TI PCM1862 to regenerate the master clock and mix in additional analog sources, with one of its stereo outputs split to all 4 i2s input lines of a PCM1681, and the other of its stereo i2s outputs back to the 96boards where supported or hackinable. I.e., the dragonboard 410c's i2s input pin is N/C, but the second channel is actually available at pin 27 (and contrary to everybody telling me initially that it is impossible for it to receive i2s input, the new hardware manual makes it VERY clear that it CAN receive). That can be used to feed a microphone signal back in... which yes, the dragonboard 410c actually has analog microphone inputs (3 of them, in fact), but not the HiKey board.
What I was hoping I would be able to find, is a one-chip audio processor that could receive one or more i2s and analog inputs, and produce at least 4 analog line outs at a high enough level to drive an automotive amplifier. Unfortunately, such a part doesn't seem to exist. ADAU1701 comes incredibly close, except that to receive i2s input in slave mode, it requires a synchronous master clock input. What this means is that I can't run it straight off a 96boards device except by first feeding that i2s through some DAC that doesn't need a master clock. In the case of ADAU1701, that would either tie up the only stereo analog input, or I would require an additional DAC that is operating in slave mode to the ADAU1701. That is obviously far too messy of a solution.