Death toll isn't the only metric for accident severity. Environmental pollution is really important too.

And with Fukushima we were kinda lucky. It was still a double meltdown with several hydrogen explosions and could have been much worse.

I think I already addressed this point. You also need to think about second-order effects, but like I said, there are second order effects to all of these solutions. Just because nuclear's side effects are more easy to dramatize doesn't mean that it is necessarily more deadly / harmful to the environment.

Death toll so far. Projections of cancer deaths from Fukushima go up into the hundreds.

That's not true.

The WHO says there will be no detectable health effects.

That's because the deaths would be lost in a sea of ordinary cancers. Hundreds of additional cancers would not be detectable; that would be below the statistical noise floor. Not being detectable does not mean they won't occur.

Anyway, let me steelman what you're aiming at here. I think you want to argue not that hundreds of deaths won't occur, but that hundreds of deaths don't matter that much. These are statistical deaths, so it's appropriate to treat them using the "statistical value of a human life". This is the value of a life to be used for policy purposes (like deciding if a safety measure is needed, if spending on a medical treatment is appropriate, etc.) In the US, it's around $12M per life. So, 200 (say) deaths would have a value of $2.4B. This is not enormous compared to overall cost of the accident, even to the utility. It could be reasonable to treat radiation releases as at Fukushima by fining the polluter by an amount related to this value.

Under this sort of regulatory regime, the purpose of regulations is not to avoid all releases, but to keep the releases small enough that the utilities would have the resources to pay the fines. So, no 100,000 death accidents. Nuclear power plants designed to this concept could allow some small radiation release in accidents.