Consensus Architecture: Separate block authoring from collation submission

[rphmeier]

Motivation:

1. Enable nodes to submit collations for blocks they haven't themselves authored (probably needed for implementing PBFT/HotStuff style consensus)
2. Disconnect the concept of "collation" from being a single block - collations may become bundles of many small blocks authored at low latency

We should investigate architectural components which separate the logic of submitting collations from the actual block-building logic in Cumulus.

Collation workers should basically just be futures that submit collations on certain events, such as new blocks authored by the local node.