Add tolerance to divisible by condition

[sovdeeth]

Problem

The divisible by expression is particularly susceptible to floating point error compounding during the modulo operation. This means many common divisions like 0.3/0.1 return false when they should be true, due to slightly fp inaccuracies.

Solution

I may have gone a bit overboard. Nearly all reasonable cases can be fixed with a check against Skript.EPSILON, which ensures the result is within 1e-10 of evenly divisible. However, this fails when working with large or small numbers, or a combination of the two. For example, 3,000,000 / (1/3) fails as the error is slightly larger than 1e-10.

Using ULP is sadly not helpful here, since the ulp for 3 million is 0.25 (way too large) and for 1/3 it is roughly 1e-17 (way too small). A fixed epsilon also fails if the user wants to work with small numbers, as a dividend of 1e-10 or smaller will always return true, no matter the divisor.

This PR implements a solution by defaulting to 1e-10 as the tolerance, but allowing the user to manually provide a tolerance if needed. This should covers nearly all cases by default and provide options for users it does not cover.

A few extra shortcut paths were added to speed things up.

Testing Completed

Additional tests were added to CondIsDivisibleBy.sk to cover the new functionality.

Supporting Information

---

Completes: #7930
Related: none