063da94bd8
This uses a bit more memory (each voice needs to hold buffers for the deinterleaved samples of each channel, instead of just one buffer for the current channel being mixed on the device), but it will allow for handling formats that need or prefer their channels decoded together.