fixMantissaOverflow should not round up

[IvanSanchez]

The current implementation of fixMantissaOverflow seems to always round up (when the first non-significant digit is 5), as noted on

notations/spec/infix/infix-eng.spec.js

Lines 14 to 15 in 3799032

	expect(notation.format(34.5, 0)).toBe("35₀"); // Is the rounding right??
	expect(notation.format(33.5, 0)).toBe("34₀"); // Is the rounding right??

I think it'd be better to implement the "round half to even" algorithm (see also the IEEE754 stuff about rounding).

In a nutshell:

A tie-breaking rule without positive/negative bias and without bias toward/away from zero is round half to even. By this convention, if the fractional part of x is 0.5, then y is the even integer nearest to x. Thus, for example, +23.5 becomes +24, as does +24.5; while −23.5 becomes −24, as does −24.5. This function minimizes the expected error when summing over rounded figures, even when the inputs are mostly positive or mostly negative.

[dan-simon]

fixMantissaOverflow isn't doing rounding in those cases, the rounding up is because 3.45 and 3.35 are rounded to one decimal place as 3.4 and 3.3 (line 58 of community/infix-abstract.ts, specifically value.mantissa.toFixed(places)). For example, 35.5 is rounded to 35, which is not rounding up, but can be explained by (3.55).toFixed(1) being 3.5.

The default formatUnder1000 (used by, for example, Scientific notation) also has this issue. This is because x.toFixed(0) in JS uses round half away from zero rather than round half to even. I'm not sure how to generalize x.toFixed(i) for i not equal to zero to use round half to even, but my best quickly thought out guess is something equivalent to if (4 * Math.pow(10, i) * x % 4 === 1) {return (x - Math.pow(10, -i) / 2).toFixed(i)} else if (4 * Math.pow(10, i) * x % 4 === 3) {return (x + Math.pow(10, -i) / 2).toFixed(i)} else {return x.toFixed(i)}. However, this might have floating point issues.