`@batch` slower than `Threads.@threads`

Tried to see if replacing `Threads.@threads` with `@batch` would improve performance on smaller workloads for a function. This seems to introduce some kind of data race (which wasn't present with `Threads.@threads`), which results in many `NaN` values and incorrect results, which were caught by the tests. The relevant loop is [here](https://github.com/ParadaCarleton/ParetoSmooth.jl/blob/main/src/ImportanceSampling.jl#L69). Not sure what the cause is; the `for` loop is looping over calculations for each data point, and each one of these should be completely independent of the calculations for every other data point.