Add Arbitrary implementation for small arrays

[tbu-]

Add Arbitrary implementation for small arrays

[tomjakubowski]

Isn't Arbitrary limited to a single implementation for "T" subject to some bounds (to avoid overlapping instances)?

[tbu-]

Yea, you're right, the second version doesn't work.

[BurntSushi]

Are there any concrete use cases for this? Could we get away with only adding impls for a few? I'm hesistant to add a bunch of impls if they increase compile times and aren't holding their weight.

[tbu-]

I currently generate arrays of bytes to test that my serialize/deserialize roundtrips.

[cyplo]

This would be extremely useful for small byte arrays for me as well, functions that I test typically have [u8; 32] as an argument type, not necessarily even a slice but a known size array.

[BurntSushi]

I'm going to call this a duplicate of #187. We've held out this long. I suppose we can hold out a little longer for const generics to land.