Redefine score curves with p90 rather than podr

[paulirish]

It's come up a few times.. most recently here #10495 (comment)

we should really consider reworking our math so we can define in terms of the 90 cutoff :) haha

It makes sense, since 90 is passing and a key threshold. Our source of truth should use them instead of numbers SO CLOSE to it.

[paulirish]

Now, the hard part (the MATH) has now been done, thanks to @brendankenny.

desmos: https://www.desmos.com/calculator/wbudrdywv2

Brendan's JS:

const INVERSE_ERFC_ONE_FIFTH = 0.90619380243682323694542901648674160242080688476562500;

function getLogNormalPercentile({median, ninetieth}, value) {
  // Shape (σ) is `|log(ninetieth/median) / (sqrt(2)*erf-1(1/5))|` and standardizedX
  // is then `1/2 erfc(log(value/median) / (sqrt(2)*σ))`, so simplify a bit.
  const xLogRatio = Math.log(value / median);
  const ninetiethLogRatio = Math.abs(Math.log(ninetieth / median));
  const standardizedX = xLogRatio * INVERSE_ERFC_ONE_FIFTH / ninetiethLogRatio;
  return (1 - erf(standardizedX)) / 2;
}

left to do:

• adopt in statistics.js w/ test
• calculate p90 for all existing curves and redefine them, confirming their shape doesn't change. set new desmos links.
• update scoring calc, i guess.