Uh, security at multiple levels, dispute resolution, dealing with inaccurate floating point math, CAP theorem limitations, throughput.... And those are just the challenges I can come up not having worked in the financial clearing domain.
No. You cannot just build one of these at a code jam, you cannot launch a startup to build one of these in a few months. It's a system with one of the highest fidelity requirements outside of medical equipment. Even space technology is allowed to fail for being off by a little bit. Financial systems at scale are hella difficult.
As I understand it, they use fixed point values in finance, so it's not any worse. You could use arbitrary precision arithmetic if performance wasn't an issue, but it is, and why would you want that anyway?
I can't really comment on scale, but it's been a couple years Russia got kicked out of SWIFT, so they've had time.
I don't know. The article quotes the Russian ambassador saying it needs to integrate with the banks and whatever systems are in place currently for each country, but that doesn't necessarily mean SWIFT as well. But even if it's a direct competitor, the possibility of future SWIFT integration has to be considered. I'd be surprised if they blanket never wanted it to happen.