refactor(core) with repo

[btc]

This PR introduces the repo to the core package. The primary objective is to share more code between the live node and the integration tests (dev/prod parity). A secondary objective is to make routing elegantly pluggable. A non-goal is to facilitate creation ephemeral instances (alternate datastores).

This is not ready for CR. There are some open items:

• who owns the repo resources? Repo or Node?
• implement offline RepoConfig
• pass sharness tests
• update all NewIpfsNode call-sites
• use the core constructor in integration tests (epictest)

[jbenet]

@maybebtc it looks like the abstraction you have in mind is something that includes network, dht, and bitswap. that may be a good abstraction to make, but I think the "Repo" should mean specifically only an ipfs node's data storage repository (no Network). think the same thing as a git repo (all the objects, refs, config, hooks, etc). A node instantiated can operate on a repo (or more!) and add/modify data. I think an IPFS repo should be:

• a Config (defining the node and many options)
• a Datastore (for blocks and other data needed to be stored locally)
• a Blockstore (wrapping the Datastore)

And an FSRepo impl should include a way to initialize it with an os path. (Another impl -- for example -- could be an S3Repo that happens to store everything in S3).

This matters because I plan to entirely rip out the merkledag + repo portions to make simple unixy tools that manipulate ipfs repos without at all touching the rest of ipfs.

[btc]

merged in #538