The Cortex chip is capable of rewriting its own flash memory so I don't see any reason that someone could not construct an SPI based bootloader that can do what is needed. Now, I don't know of any off-the-shelf already done version of this but there should be plenty of examples of bootloaders that could be modified. This would then require zero hardware changes - just the addition of a bootloader located somewhere outside of where the programming must happen. This will probably badly break Arduino compatibility, at least for program uploading, but such is the price you pay.