Merge lp://staging/~jderose/filestore/helpers into lp://staging/filestore

This merge:

* Adds 3 new highish-level FileStore methods:

  FileStore.content_md5(_id) - compute a known correct MD5 hash (for uploading to S3 using boto)

  FileStore.verify_and_move(tmp_fp, _id) - check content hash of a file in .dmedia/partial/, and if correct, move the file into its canonical location. This replaces an equivalent method of the legacy FileStore that is used by the S3 and BitTorrent backends.

  FileStore.hash_and_move(tmp_fp) - calculate content hash of a file in .dmedia/tmp/, and then move the file into its canonical location. This replaces an equivalent method of the legacy FileStore that is used by the transcoder.

* Makes the use of the `Leaf` namedtuple consistent. Experience has shown the code is simpler and more robust if you keep track of the position in the file (leaf_index) a given chunk of leaf_data corresponds to. As of several revisions ago, when a leaf is read by reader() or batch_reader(), the index and data are bundled together right there in a Leaf(index, data) namedtuple. But this merge fixes some inconsistencies higher up:

  - Consumers of `Leaf` no longer unpack it into (i, data)

  - Consumers of `Leaf` yield/return the exact `Leaf` rather than (i, data)

  - Hasher.update() has been renamed to the clearer Hasher.hash_leaf()

  - Hasher.hash_leaf() now takes a `Leaf` instead of `bytes`

  - Hasher is now more robust as it makes sure the Leaf.index from the readers matches up with its expected leaf_index