In this PR we're doing several breaking changes:

• move all detailed implementations to std/util/math
• make all Math's methods polymorphic
• remove non-standard NativeMath#scalebn
• remove non-standard NativeMath#rem
• remove non-standard NativeMath#mod (but not tests)
• remove NativeMathf completely
• for now, we have only single Math.random (for f64 type)

Caveats:

1. I have observed quite often the script that users from JS for example:

function foo(x: i32, y: i32): i32 {
   let res = Math.floor(x / y) as i32;
   return res;
}

Before this PR this will create unnecessary cast to f64, Math.floor and downcast to i32. With this PR this code now equivalent to x / y without all this unnecessary work due to Math.floor, Math.trunc and etc accept integers and pass through it "as is".

2. Now following code will generate a compilation error:

const cos3 = Math.cos(3);
// -> ERROR: Math.cos accept only f32 or f64 types
// Solution: it may be fixed as Math.cos(3.0)

const x: i32 = 123;
const sqr = Math.sqrt(x);
// -> ERROR: Math.sqrt accept only f32 or f64 types
// Solution: Math.sqrt(x as f64)

This is the most painful caveat, which can lead to many breaking changes in some code.
Perhaps we could add new generic builtin helper which forces parameter type to float. Something like:

// builtin will be equivalent to this:
type OnlyFloat<T> = T extends f32 | f64 ? T : f64

function sqrt<T extends i32 | i64 | f32 | f64>(x: T): OnlyFloat<T> {
   if (isInteger<T>()) {
     return builtin_sqrt<f64>(x as f64) as OnlyFloat<T>;
   }
   if (isFloat<T>()) {
     return builtin_sqrt<T>(x) as OnlyFloat<T>;
   }
}

// now we can use integer inputs which will forced to f64 for outputs but safely preserve f32
const sqrt3 = sqrt(3); // infer as sqrt<i32>(x) -> f64

Thoughts?

• I've read the contributing guidelines
• I've added my name and email to the NOTICE file