Roofing is a dangerous profession, so rooftop solar is probably much more dangerous than other solar.

You need more solar panels to get the same amount of energy. So since all the practical risk factors in nuclear apply to solar, and solar requires more activity, we'd expect solar to have more deaths per energy unit.

That said, neither nuclear or solar really kills anyone per TWh, so solar might score better by statistical fluke since the variances are so low.

I personally suspect that we don't count sunburn -> skin cancer deaths from solar. If we did, the radiation risks would ironically be higher for it than for nuclear power. Still basically 0 though.

If you add the solar at the time of building the roof you'd probably reduce this dramatically, as the house would probably be shrouded in scaffolding.

Yep, which to elaborate on your point means we could expect that number to be lower if it covered all solar.