Signed int keys order

[maurolacy]

Closes #489.

There are a number of different ways to do the signed int key modification (a number of different impls in the commits). Not sure what is the best one; opted for doing it without mutable slices, and without branching.

I guess the only way to know for sure would be by benchmarking them. Will create an issue for it, and we can do in the future.

[ueco-jb]

Good to see that this idea actually worked just fine.

[ethanfrey]

looks good.
I would clean up the test cases to be more explicit rather than just contain the same math being tested

also, let's add test case that -321.to_cw_bytes() < 0.to_cw_bytes() < 652.to_cw_bytes()

[ueco-jb]

lgtm
Personally I'd replace couple big numbers with constants for better transparency, but overall it's fine.