caniuse usage data for Firefox is suspect

[foolip]

Splitting #3 (comment) into its own issue:

https://gist.github.com/foolip/87088161970e9cc152cc8b425f8474ca is a dump of the caniuse data for Firefox usage from 2023-05-22T09:14:47.000Z, which is what ends up being used to compute the 83.08% number in the previous comment.

The data seems implausible, claiming that 0.48% of global usage is still on Firefox 1-71. What's most implausible is that the usage is roughly even spread across those versions. Firefox 2 supposedly has 0.004118%, and 71 has 0.004425%. @jgraham @zcorpan does this seem remotely plausible to you?

One meta conclusion here is that the analysis is extremely sensitive to the shape of the long tail of version decay. I don't know why the caniuse data (from statcounter) looks like this, but it's also a lesson if we try to model version decay. In particular, it's importan that the cumulative usage of old versions is roughly correct. I can't imagine that it is for Firefox in this data.

[foolip]

Taking it upstream as well in Fyrd/caniuse#6733.

[foolip]

A bug was found and fixed in Fyrd/caniuse@87ff4c1. The data still seems odd, I'm poking more.