Merge lp://staging/~jderose/filestore/dbase32 into lp://staging/filestore

For details, see this bug:

https://bugs.launchpad.net/filestore/+bug/1130156

Changes include:

* Use compatibility functions from `dbase32.rfc3548` so that it will be very easy to switch to Dbase32, in particular use `isb32()` instead of set operations with B32ALPHABET; we can just replace with `isdb32()` when needed

* Removed protocol selection-by-ID-length functionality, for now; this was incomplete and as we have a demanding migration ahead of us, I backed this feature out so we have less complexity to deal with during the migration

* `Protocol.hash_root()` now returns a binary digest instead of a Base32 encoded ID, and `filestore.protocols` is now completely encoding agnostic, does nothing with encoding itself

* `Protocol.hash_leaf()` now transforms *leaf_index* into the key bytes rather than relying on `Protocol._hash_leaf()` to do this

* Likewise, `Protocol.hash_root()` now transforms *file_size* into the key bytes rather than relying on `Protocol._hash_root()` to do this

* `Protocol` is now a base class without any hashing implementation itself; real implementations are now in the `SkeinProtocol` and `OldSkeinProtocol` subclasses

* Changed V1 to use a 240-bit digest and DBase32 encoding, updated test vectors accordingly

* Split old test-vectors.json file (which contained both V0 and V1) into separate V0.json and V1.json files

* Added MD5SUMS.json with md5sums of various values used by the test vectors; these are helpful for debugging an implementation, as mention in the specification

* Huge refactor of the unit tests for `filestore.protocols`; in particular, no digest values are in the unit tests anymore; instead, digests come from the JSON data files only, making it much easier to tweak a protocol without having to make huge changes in the unit tests

* misc.build_vectors() now takes optional *encoder* kwarg, which defaults to `db32enc`, so the test vector building is no longer hard-coded to a specific encoding