Unit tests are tautological in many cases

[pavelkomarov]

There are a lot of unit tests that just call a function and make sure the output matches some expected vector, but the values in that expected vector are awfully specific, so I suspect they were generated from the code itself. Ideally we take the derivative of a suite of known functions (maybe plus some noise) and check that the result is close enough to known derivatives, so the tests are independently correct. I do this in spectral-derivatives (see my lambdas at the top), and @andgoldschmidt and @Jacob-Stevens-Haas did this in derivatives (see funcs_and_derivatives). I'll implement something similar for this repo.

[pavelkomarov]

I think some tautologies remain in the optimizer tests, but those tests aren't quite of the same flavor as tests of the basic differentiation functions, being a bit more opaque, harder to specify the definitely-right choice ahead of time. That's its own kettle of fish. I'll close this one, because as of #113 all the differentiation methods' tests go through test_diff_methods.py, which supports a neat --plot flag to show how methods are doing against tests functions, with and without noise.