Will be interesting to see if there is an official answer to this, but it's been something I've been meaning to experiment with.
One thought is master vs. slave mode. For the MT7688 to generate the clocks, the driver would need to put it in master mode, if it is in slave mode, then it would expect that the clocks would be generated by an external CODEC. Given the comments that have been made about the fractional-N clock generator, that may be the preferred way to operate if you have something that can generate clocks externally.
Another thing I've contemplated trying is to momentarily ignore the driver and just use the
devmem tool to directly set the register bits per the MT7688 data sheet to enable and configure I2S clocks, to at least see what the hardware does.
Would be great if there is an official "do this, this, and this" procedure for I2S audio input through Linux though.