Better testing

[valentinsulzer]

Currently most tests are of the form

@test a ≈ b atol=0.01

which uses the L2-norm and so can give errors even if the vectors are equal within tol elementwise (L-inf norm). (Linf(x) <= L2(x) <= sqrt(n)Linf(x))
Should we be testing instead

@test all(isapprox.(a, b, rtol = 0.01))

? Or do we really want to test the L2 norm?

See https://discourse.julialang.org/t/unit-test-equality-for-arrays/8138/13

[ChrisRackauckas]

Element-wise testing is probably better.

[valentinsulzer]

So using that isapprox? Or is there a cleaner way?

[ChrisRackauckas]

Broadcasted isapprox is the right idea.