Optimize NativeMath/NativeMathf to use WASM builtins where beneficial

[dcodeIO]

For example:

• There are some places where it appears to make sense to use the copysign instruction instead of using a branch (might need a benchmark)
• JavaScript's Math.round always rounds exact fractions of 0.5 towards +Infinity, even if negative, and the code currently uses a branch for that.
• Possibly quite a few more things I haven't yet thought of.

See: std/assembly/math.ts

[rdosanjh]

Could i take this?

[dcodeIO]

Absolutely, there is nobody else working on this as far as I know, and there are plenty of libc-test cases already that should help :)