There is no way this was safe before.

Be careful to unlock on the error paths.

No code attempts to open the same repo multiple times.

Error handling is still buggy, but it's starting to get clearer.

repo.Once now does refcounts on a whole Repo level, returning the same
pointer to multiple Openers. This removes the need for the weird model
of separate FSRepo instances pointing to the same underlying storage,
and the races caused by that.

This will replace the elaborate refcounting in fsrepo, to make it
easier to maintain.

Nobody calls At without immediately calling Open.

First step, a mechanical transformation. Cleanups will follow.

This is only called from `ipfs` command line tool well before it opens
the repo. The behavior change here causes a false positive if the
current process has already opened the repo. That's a bit late to ask
this question, anyway, and is not expected to have ever triggered.

The only caller is `ipfs init`, which at that time does not hold a
repository open, and refuses to run on existing repos anyway.

Dead code cleanup

rm util.NewByteChanReader()

Dirty hack to fix race conditions in the daemon

Ipns/refactor dagmod

Nothing uses it.

Nothing uses it, and offset+limit is a bad query mechanism for
mutating data.

The version actually used is in fsrepo, and belongs there as it knows
about fsrepo file layout.

Repo cleanup

t0060: reactivate daemon tests

ipfs object put: return error if object is empty

sharness: add function to test curl response