Thank you for this!

I was just doing some playing around here and I believe there are some other edge cases too.
As an example Decimal::MAX (as f64) and Decimal::MAX - 1 will both also fail since the current max for the f64 parsing algorithm is 79_228_162_514_264_337_593_543_950_330.

That said, I think this is taking things in the right direction so will merge this in.

@paupino AFAICT those problems don't exist since f64 can't represent those values. My test covers three adjacent f64 values, found via the unstable f64::next_down.

In the test you considered adding, the value loses precision before converting to decimal.

@finnbear Oh cool - I wasn't aware of that function. A handy one to have in the toolkit - particularly for testing floats!