Amusingly, this theory (but not this implementation) actually might work. It comes down to the host board and NVMe controller.
If we assume a well-behaved host board, then at power loss the NVMe should simply halt all operations for want of instructions. So all the battery is actually doing is preserving cache state. A well behaved host should command cache flush at initialization, and a well behaved controller should have a 'timeout' that flushes cache to disk if no instructions are received in a set period of time.
Similar concept to what AMI's been doing for decades with MegaRAID, except MegaRAIDs only preserve cache (for obvious reasons.)
But I doubt very much that's an actual supercapacitor, or one with enough capacity to handle anything longer than a brief power blip. Hell, to be entirely honest, that looks like it's just a CR2032 case. And it's probably not even actually hooked up.