It's likely you are facing one or possibly the combination of two known issues.
One is that because an incompatible 32 megabyte SPI flash was used, accessing the whole of it requires putting it in a mode incompatible with boot, so unless something resets the flash chip when the CPU does, a warm boot will fail. Most other vendors avoided this by operating their 32-megabyte flash in 4-byte address mode all the time (but that would have required different strapping for the 16 meg and 32 meg versions).
Another is that if you have anything connected, say a debug serial port, the data lines itself will keep the MT7688 alive enough that it won't do a clean power on reset . Potentially in this case you have a similar leakage path where an I/O is keeping your flash chip alive enough during the power interruption that it isn't losing the boot-incompatible 4-byte address mode state. Or you could have residual charge on a capacitor.