The 2% number is an overestimate. For the US over two years we had about 800K dead and 250M infected (ballpark assuming that 80% holds everywhere), which yields about .3% death probability per individual ever infected. For context, a bit less than 1% of the entire population dies each year (2,854,838 died in the US in 2019 https://www.cdc.gov/nchs/products/databriefs/db395.htm#secti...), so about 6M people would have died in two years anyway even without covid.
You are dividing confirmed covid deaths by estimated covid infections, and that's not really valid.
When researchers study those with natural immunity, they examine people who are confirmed to have had covid before, so the CFR is the right number to use.
BTW, In the US covid has a CFR of 2%. It is not an "overestimate." It's not even an estimate. It's just what the number is. Obviously that number changes depending on the time frame, and it is (thankfully) dropping.