One really useful resource here will be this burnup table (https://www.eia.gov/nuclear/spent_fuel/ussnftab3.php), which shows that in 2017, there were 3482 PWR assemblies discharged, initially containing 1526.5 tonnes of uranium, and that on average the assemblies were burned to 46.2 MWd/kgHM. (where HM mean heavy metal, and is always referring to the mass of initial uranium of the fresh assembly)

This means that a single average 2017 PWR assembly gave out 21.5 GW-days of (thermal) energy.

(I rely heavily on the GNU Units command line program for all my conversions, I'll paste the inputs in here).

$ units "1526.5 tonnes * 46.2 MW*day/kg/3282" "GW*d"
* 21.488208

PWRs are about 32% thermally efficient, so those thermal GW-days are roughly equivalent to 6.9 GW-days of electricity, or 1.7e8 kWh.

$ units "1526.5 tonnes * 46.2 MW*day/kg/3282 * 0.32" "kW*hour"
* 1.6502944e+08

The dry casks at SONGS hold 24 assemblies per cask (will be more later). I assume Diablo is similar, but not totally sure.

So a dry cask of PWR fuel represents about 4e9 kWh of electricity.

$ units "1526.5 tonnes * 46.2 MW*day/kg/3282 * 0.32 * 24" "kW*hour"
* 3.9607066e+09

As you found, each kWh of nuclear electricity saves (490-12) = 478 grams of CO₂ (equivalent) over gas.

Thus, a single dry cask represents a savings of 1.9 million tonnes of CO₂-eq!!

$ units "1526.5 tonnes * 46.2 MW*day/kg/3282 * 0.32 * 24 * ((490-12) grams/(kW*hour))" "million tonnes"
* 1.8932177

WOAH! This is actually a really cool comparison!! Thanks for the idea.

Oh and btw, the rule-of-thumb number of a tonne of fuel for a GW-yr is indeed wrt the atoms that fissioned, which in a normal reactor is mostly U-235.