(test was failing)

Erroring out in core setup should cancel the context
to ensure subsystems are shut down. This has to happen
all over the place we use contexts.

@perfmode @whyrusleeping

**For now**, we don't need to load/parse the private key
(which causes a signficant delay in commands) when doing
things entirely offline. This may change, and in that case
the private key should be loaded on demand.

See https://github.com/atom/atom/issues/3660

wish go-dep warned you about dependency problems.

cc @whyrusleeping

cc @mildred

Rework config file location

Removed config.Filename and replace it with config.GetConfigFilePath that
takes a configuration directory as argument instead. Makes code simpler.
ipfs.getConfigDir now also return the default configuration dir instead of
an empty string in some cases.

Peerstore Fix

it's failing when machine slow (travis)

This is to avoid having to import ds elsewhere.

config: rename addresses

Rather than pushing errors back down to lower layers, propagate the
errors upward.

This commit adds a `ReceiveError` method to BitSwap's network receiver.

Still TODO: rm the error return value from:

    net.service.handler.HandleMessage

This is inspired by delegation patterns in found in the wild.