Ruck

**Additions**
- Functionally equivalent version of NSGA-ii implemented at `ruck.functional.select_nsga2`
+ if `numba` is installed then the function will be JIT compiled for up to 65x faster performance
+ minor differences exist compared to `ruck.external.deap.select_nsga2`, but overall results should be similar.

**Deprecations**
- `ruck.external.deap.select_nsga2` has been deprecated and will be removed in `v0.3.0`

**Additions**
- Added `MANIFEST.in`

Breaking change if you are using the ray helper functions that allows us to remove the optional dependencies. The core ruck API remains the same.
- `ruck.util._ray` moved to `ruck.external.ray`
- added `ruck.external.deap` with `select_nsga2` that uses deep behind the scenes. The goal is to implement the ourselves in the near future. Currently it is a bit slow, and means me need extra deps.
- More examples!

Changes
- Allow tuples of floats as fitness values

Changes

- Ray remote function support was broken because of algorithm design decisions and use of nested functions and decorators. Updated to remove these limitations. API changes slightly.
- `R.mate_crossover_nd` added to compliment `R.mate_crossover_1d`
- Training checks that all the values in a population maintain the same type, this can help avoid ray multithreading errors when object refs are required instead of values.

Initial Release