Faster custom losses

[MilesCranmer]

This makes custom losses faster by changing

sum(loss.(x, y))

into

l(i) = loss(x[i], y[i])
sum(l, eachindex(x))

It's a small change but improves the speed of custom losses quite a bit (because sum(::Function, ...) performs the sum with a map reduce)

[github-actions]

Benchmark Results

	master	9418230...	t[master]/t[9418230...]
search/multithreading	54.7 ± 2.4 s	52.6 ± 1.5 s	1.04
search/serial	59.2 ± 0.91 s	58 ± 0.86 s	1.02
time_to_load	16.8 ± 0.083 s	17 ± 0.17 s	0.993

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).