Enhance comparison types to support bigint comparison

[Emiyaaaaa]

• fixed Numeric comparison types: support for bigints? #1413

[sindresorhus]

1. The nonliteral bigint fallback runs before the infinity handling, so comparisons where one side is broad bigint and the other side is PositiveInfinity/NegativeInfinity widen to boolean, even though the answer is fixed. For example, GreaterThan<PositiveInfinity, bigint> should be true, GreaterThan<bigint, PositiveInfinity> should be false, and LessThan<bigint, PositiveInfinity> should be true, but they all currently come out as boolean. Moving the broad bigint fallback until after the infinity cases should keep the existing precision.

2. The tests add mixed bigint/number support, but that path is not correct for decimal number literals. For example, GreaterThan<2n, 1.5> currently resolves to false, and GreaterThan<1.5, 2n> resolves to true, because the positive numeric string comparison treats "1.5" as longer than "2". The issue only asked for bigint-to-bigint support, so the simplest fix is probaly to drop mixed number/bigint support from this PR, or make the contract explicitly integer-only if that is intended.

Also the public docs still only talk aboutt number and non-literal number, so they should be updated to mention bigint and broad bigint behavior too

[Emiyaaaaa]

DONE

[som-sm]

Umm... GreaterThan<2n, 1.5> still returns false, and GreaterThan<1.5, 2n> still returns true?

[Emiyaaaaa]

Umm... GreaterThan<2n, 1.5> still returns false, and GreaterThan<1.5, 2n> still returns true?

GreaterThan previously only supported integers; I have updated the type description.

[som-sm]

Add tests for PositiveInfinity/NegativeInfinity and number.

[som-sm]

I guess it's fine to return boolean in all these cases. PositiveInfinity and NegativeInfinity are anyway not that common, and returning boolean isn't inaccurate, it's just imprecise.

Also, if we handle these cases, then GreaterThan<PositiveInfinity, number> should also return false, which it currently doesn't (returns boolean currently).

[som-sm]

I think it makes sense to keep the implementation simple, and just return bigint if bigint extends A | B. And remove the IsFinite type.

[Emiyaaaaa]

@sindresorhus