fix(eventlog) initialization

[btc]

This PR fixes two issues.

daemon --init breaks eventlog configuration

Introducing daemon --init broke eventlog configuration. Here's why:

1. Now, users may run daemon without an initialized repo.
2. the configureLogger hook checks to see whether the command usesConfig and usesRepo. If both are true, the logger is configured to use the IPFS_DIR. This is necessary since the logger logs to a file.

Since the daemon doesn't require a repo/config and configureLogger only runs if the upcoming command uses the config/repo, logs are not configured before running the daemon.

Notice that the one actual requirement for the hook is the existence of a config/repo.

Therefore, we may want to simply check this property directly. Thoughts?

runtime logging system is not configured upon initialization

Now, after initialization, we apply the config.Log settings to the logging system. This ensures logging works when the daemon is started using daemon --init

Third, this paves the way for the repo package described in #376.

[jbenet]

LGTM! 👍 I like the repo package idea. We could even move the entire config into that, and have core.IpfsNode hold a repo object:

type Repo struct {
  Path string
  Config config.Config
  Datastore ds.Datastore
  ...
}

// move the file lock stuff here.
func (r *Repo) Lock() {}
func (r *Repo) Unlock() {}

Or something?

Also, maybe Repo should be an interface. What if we had in-memory only repos? or repos not on the filesystem?

type Repo interface {
  Config() config.Config
  Datastore() ds.Datastore
  ...
}

type fsRepo struct {
  path string
  ...
}

func NewFSRepo(path string) (Repo, error) { ... }

And we could move a lot of the config read/writing into the repo implementations. thoughts?

[btc]

RFM

re repo: hell yes (replied in #376)