Roadmap Link: storj/roadmap#12

The goal here is to refactor upload (and later download) to reduce the usage of buffer allocations.
In the current model, we use io.Copy and buffer-based approaches to copy data between different abstractions.

The goal here is to rethink the abstraction to do as few buffer copy as possible.

As a side-effect: with better abstraction, we can separate the different functions easier: it can be possible to do encryption locally but calculate EC and connect to storage nodes from a lightweight gateway service.

A possible approach is evaluated in the https://github.com/storj/uplink/tree/uploadng branch.

This is not a fully working solution, just a good skeleton that contains all the important parts. Initial performance tests (based on unit tests) show that it can be significantly faster.

However, the full implementation requires either further improvements on the abstraction or simplification in the protocol change, as not all the encoding information is available in advance...