The new data transfer

[hannahhoward]

Goals

Support many things in data transfer, including:

• pausing and resuming
• piggy backing on graphsync for payment protocols like retrieval (via intermedia vouchers and the like)
• actually being able to potentially support protocols other than graphsync
• potentially support pause/resume across restarts
• support pausing and resuming from either side

Implementation

There are a number of changes here, but the basic architecture decisions are:

1. Factor out transport specific details from the machinery of data transfer itself -- which handles tracking tranfers, managing states of channels (including pausing and resuming, sending vouchers around). This is still a bit WIP but there's a much greater separation, which enables the non-transport specific implementation to focus on the significantly more complicated logic of allowing not simply one-and-done transfers but transfers that can stop and start, and have many intermediate points in the middle.
2. Move channel tracking to the oft-used FSM architecture we've used for state tracking elsewhere - while perhaps we've used this hammer so much everything looks like a nail, the ability to easily specify state transition logic makes it simply to track what's happening when and have more complciated sets of steps.
3. Actually unit test this library, then write stress test level integration tests. In the initial verison of the library, in part because separation between the transport and transfer logic was minimal, most of our tests were effectively integration tests, but at the same time they weren't testing all that much. The new version leans heavily on unit testing, but also includes integration tests that test a lot more than the old version did.

[ingar]

👍 as per our discussion