fix issues 411,Reason: NewFromFloat returns the shortest decimal for round-trip.

[8xmx8]

Perform an exact decimal expansion strictly based on the true binary floating-point value (rather than the shortest round-trip representation).,see：#411

[juanjoseluisgarcia]

why this PR has not been merged yet?

[mwoss]

Thanks for opening and PR and helping to fix #411.

It looks great, however, this changes the public contract of NewFromFloat / NewFromFloat32 from "shortest round-trip representation" to "exact binary value," which is a breaking behavioral change.

Examples:

On master:

NewFromFloat(0.1).String() # "0.1"
NewFromFloat(0.3).String() # "0.3"

On this branch:

NewFromFloat(0.1).String() # "0.1000000000000000055511151231257827021181583404541015625"
NewFromFloat(0.3).String() # "0.299999999999999988897769753748434595763683319091796875"

Anyone comparing .String() output, persisting, etc, these values will be affected. The "shortest" behavior is the long-standing documented contract and is what most callers may rely on.

So maybe we could fix that narrower case from #411? I meant exact expansion only when the float has no fractional part while preserving shortest round-trip for fractional values?

If we would like to fix both, I would implement new public method like NewFromFloatExact