Test machinery for implicit functions

[isovector]

This PR adds quickcheck machinery for testing implicit functions. It:

1. Adds Arbitrary generators using the public interface, so that we can generate random symbolic objects.
2. Adds an Observe instance for each, saying that two symbolic objects are equal if their getImplicit functions agree at all points. Future work should also ensure they have equal getBoxes, but due to rotate3 being currently broken, this isn't true today.
3. Uses the above to write property tests (equivalent to 1600 unit tests!) for rotation.

The property tests have some unhappy results; getBox (rotate3 z) is broken, and that rotate3 (x, y, z) for x, y, z not in {0, 1} doesn't seem to satisfy any compositional properties. Fortunately these should both be fixed in #314.

There's also quite a lot of implicit bug workarounds in the isValid functions, necessary for working around partiality (like #304 and #306) in the library.

[julialongtin]

because of the nature of floating point, anything that looks at decimal position is going to be fragile. there's functions for taking the 'integer' part and the 'multiplier' part and examining those seperately, for cases like this.

[julialongtin]

https://hackage.haskell.org/package/ieee-0.7/docs/Numeric-IEEE.html

take a look at sameSignificandBits . i think. I'm good with merging this without this, it seems good enough already.

[isovector]

I'd prefer to merge this as is, but wouldn't push back too strongly if you're keen on it.

[julialongtin]

And just a general point: the addition of another test system means we should probably have a TESTING.md file, describing all of the different test systems.