Combining tests and benchmarks

When trying to incrementally optimize and benchmarking a function, the benchmark and the regression test often use very similar code.

```elm
hashTriangleTest =
let
toList ( a, b, c ) =
[ a, b, c ]
in
Test.fuzz fuzzTriangle "hash triangle behaves as before" <|
\triangle ->
AdjacencyList.hashTriangle triangle
|> toList
|> Expect.equal (hashTriangleOld triangle)

hashTriangleBenchmark =
describe "SweepHull"
[ -- nest as many descriptions as you like
Benchmark.compare "sharesEdge shared"
"old"
(\_ -> AdjacencyList.hashTriangleOld triangle)
"new"
(\_ -> AdjacencyList.hashTriangle triangle)
]
```
Besides that fuzzing can find performance bottlenecks that would otherwise be missed (see also #3),
testing the benchmark (or benchmarking the test) gives free tests/benchmarks. It has happened to me that I made a mistake in the benchmark code and performance was looking way better than it actually was.

For completeness, here is my message from the slack

> I've been incrementally optimizing an algorithm (for delaunay triangulation) using elm-benchmark and have some feedback.
>
> My typical workflow for improving a function looks something like
>
> * write an improved (hopefully faster) version of the same function
> * write a test (often fuzz test) to check your new implementation is equivalent
> * write a benchmark to check that the new implementation is actually faster
> * remove the old code
> * remove the equivalency test (or maybe put the old version in the tests)
> * remove the benchmark
>
> Most of these steps are tedious and repetitive.
> Editor integration/code generation could help make this better, but another improvement would be to combine the benchmark and the test.
> Additionally, for more complex functions, the performance between easy and difficult inputs can vary a lot, so benchmarking on a diverse set of inputs gives more accurate results. Is this something you've thought about?
>
> Some other thoughts:
>
> * are the primitives available to write non-micro benchmarks?
> * UI that can start/cancel a particular benchmark (I'm not a big fan of running on pageload)