Refactor registry and encoding

[hannahhoward]

Goals

We're about to add a bunch more voucher types, and it seems like an important time to clean up some aspects of our protocol. This addresses various issues:

• removing the need for seperate FromBytes/ToBytes methods on vouchers on top of existing CBOR encoding/decoding methods
• making registry of vouchers thread safe
• making our protocol a single valid CBOR data-structure via cbg.Deferred

Implementation

• Build an encoder/decoder library to seperate logic around encoding and support:
  • cbor-gen types
  • ipld prime types
  • old school go-ipld-cbor types
• Build a seperate registry that handles
  • maintaining the map from TypeIdentifier -> Type, Validator (actually the registry stories a generic Processor, as several additional registries are incoming for an upcoming data transfer add)
  • keeping it thread safe
• Move encoding and decoding inside the message data structure
  • The message is responsible for encoding vouchers and selectors
  • The message receives a decoder to get back to the voucher, but it's called inside the message data itself
  • This allows us to insure we're working with CBOR encodable objects, and then to switch to cbg.Deferred inside the message
  • This is important so the whole message is a valid IPLD encoded CBOR data structure

[shannonwells]

I approve this PR

[shannonwells]

❤️

[hannahhoward]

@shannonwells now that you've turned me on to them... I am all about those table tests.