Unify and fix approaches in dtoa & strtod

[MaxGraey]

We have some misprisions and failed tests for #768 and mentioned in #337. dtoa and strtod should utilize same big float point math which could be shared. Refactor and fix this

[MaxGraey]

Currently we use Grisu2 algorithm which precise in 99.5% cases and require fallback to Dragon4 algorithm for rest of cases. I guess better investigate time to new Ryū algorithm which faster and always precise but require 10kb lookup table which possible avoid for shrinked >= 1 optimization cases

[MaxGraey]

Also new great paper about father research Ryu alghoritm and integrate it to printf

[dcodeIO]

@MaxGraey Do you think #337 can be merged into this one?

[MaxGraey]

yes, sure

[MaxGraey]

Instead Ryu we could implement more novel and faster Schubfach algorithm:
https://drive.google.com/file/d/1luHhyQF9zKlM8yJ1nebU0OgVYhfC6CBN/view

or Dragonbox:
https://github.com/abolz/Drachennest