[idea] Node fusing for speeding up execution on systems like Ray, Dask

[skrawcz]

Is your feature request related to a problem? Please describe.
For delegating to systems like Ray, it could make sense to "fuse" nodes together to reduce serialization costs.

Describe the solution you'd like
ideas:

1. We need some concept to augment the DAG.
2. We then need some pluggable way to change this logic. E.g. heuristics, vs some multi-pass logic.

Describe alternatives you've considered
Ideas:

1. Make people write larger functions. But we wouldn't do this, since it goes against Hamilton's ideals.
2. Have people tag functions that could be grouped -- seems like a good backdoor capability to have -- could work well with some more automated solution to override whatever it tries to do.

Additional context
I thought of this idea because people were complaining about Hamilton on Ray being slow.

[elijahbenizzy]

Third idea -- we can group by module...

[elijahbenizzy]

We are moving repositories! Please see the new version of this issue at DAGWorks-Inc/hamilton#45. Also, please give us a star/update any of your internal links.

Note that everything else (slack community, pypi packages, etc...) will not change at all.