Porting of efficent parsing algorithms for Double and Float

[plokhotnyuk]

https://www.reddit.com/r/rust/comments/a6j5j1/making_rust_float_parsing_fast_and_correct/

[plokhotnyuk]

The Fast Path and Moderate Path algorithms were adopted for float and double types in the 0.50.1 release. It allows parsing of up to 19-digits JSON numbers with zero allocations and less than 150 cycles of CPU.

Remaining part is Slow Path which is a fallback to standard Java parser now.

The proposed option is porting from Rust bhcomp that is faster for all inputs than any other algorithm due using a comparison between the mantissa digits and the halfway-point for the string-to-float parser.

[plokhotnyuk]

Another option is porting from C the Slow Path part of the yy algorithm.

[mijicd]

I'd like to try this one.

[plokhotnyuk]

Yet another option is considering Rust or C++ versions of the parse_long_mantissa function from the great work of Daniel Lemire "Number Parsing at a Gigabyte per Second"