I haven't seen any evidence that the actual death rate has changed at all.
See this Q&A: Does COVID-19 have a case fatality rate of 41%? Is this formula correct?
The ratio of "deaths/(recoveries+deaths)" does not describe the actual case fatality rate
Early in the pandemic, when there are lots of new people infected, almost all of the cases were unresolved because most of them were new and it takes a long time to get recovered. Imagine two people are sick, and one dies after 1 week and the other survives but is still sick for another 2 weeks. For those 2 weeks you would find, based on this formula, that the death rate is 100%, because the other case hasn't recovered yet.
Recoveries are not important to health care
In the middle of a severe situation, health care workers are most interested in finding and treating the people who are sick. Sure, it might feel good to be able to mark some people as recovered, but for those who recover on their own at home outside of a hospital, it's a low priority to follow them to find that they have recovered. Therefore, recovery numbers can lag. Some countries, like the UK, don't even report recoveries, so their patients can't be included in the "recovered" category, and if you did the calculation for the UK you'd find 100% deaths.
Only confirmed cases are counted by worldometers
Since the beginning of the outbreak, we've learned there are substantial numbers of asymptomatic people. Because not everyone gets tested, there are lots of missing people who had limited symptoms and recovered easily on their own.
In summary, the actual death rate has probably not changed
There's no secret vaccine, no evidence of mutation making it less severe, and nowhere near enough infected people for herd immunity yet.
It's also nowhere near the 15% or 21% you are citing and never was, except in the oldest and most vulnerable patients.
Note that the source you cite explains some of these things here: https://www.worldometers.info/coronavirus/coronavirus-death-rate/