Enable test runners to produce structured diff

Somewhere we (publicly) define the `ElmValue` custom type:

```elm

type ElmValue
= Record (List ( String, ElmValue ))
| CustomType String (List ElmValue)
| Int_ Int
| Float_ Float
| String_ String
| Char_ Char
| List_ (List ElmValue)
| Array_ (Array ElmValue)
| Dict_ (Dict ElmValue ElmValue)
| Set_ (Set ElmValue)

```

and then change the signature of `Test.Runner.Failure.Reason`

```diff
type Reason
= Custom
- | Equality String String
+ | Equality ElmValue ElmValue
| Comparison String String
- -- Expected, actual, (index of problem, expected element, actual element)
- | ListDiff (List String) (List String)
- {- I don't think we need to show the diff twice with + and - reversed. Just show it after the main vertical bar.
- "Extra" and "missing" are relative to the actual value.
- -}
- | CollectionDiff
- { expected : String
- , actual : String
- , extra : List String
- , missing : List String
- }
| TODO
| Invalid InvalidReason
```

which opens up the possibility of test runners producing structured diffs. We can provide a function

```elm
stringifyValue : ElmValue -> String
```

which a test runner could use if it did not want to rework its diff logic (we provide this function to help minimise disruptions)

---

related to #107