Improved efficiency and removal of Core_kernel dependency

• Improve the efficiency of the Git backend
• Expose a cleaner API for the Unix backends
• Expose a cleaner public API
• Rename Origin into Task and use it pervasively through the API
• Expose a high-level REST API over HTTP (#80)
• Fix the Git backend to stop constantly overwrite .git/HEAD (#76)
• Add a limit on concurrently open files (#93, #75)
• Add remove_rec to remove directories (#74, #85)
• Remove dependency to core_kernel (#22, #81)
• Remove dependency to cryptokit andsha1and usenocrypto` instead
• Remove dependency to caml4
• Fix writing contents at the root of the store (#73)
• More efficient synchronization protocol between Irmin stores (#11)

Improve View.merge_path + dump graphs improvement

• Support backend specific protocols for push/pull
• The Irmin Git backend can now sync with remote Git repositories
• Simplify the organisation of the libraries: irmin, irmin.backend,
  irmin.server and irmin.unix (check how the example are compiled)
• Small refactoring to ease the use of the API. Now use open Irmin_unix
  at the top of your file and use less functor in your code (again,
  check the examples)

Support backend-specific push/pull protocols

• Support backend specific protocols for push/pull
• The Irmin Git backend can now sync with remote Git repositories
• Simplify the organisation of the libraries: irmin, irmin.backend,
  irmin.server and irmin.unix (check how the example are compiled)
• Small refactoring to ease the use of the API. Now use open Irmin_unix
  at the top of your file and use less functor in your code (again,
  check the examples)

Javascript graphs & fix IrminMemory.create to create a shared store

• Fix the behavior of IrminMemory.Make to return an hanlder to a
  shared datastore instead of creating a fresh one. Add
  IrminMemory.Fresh to return a fresh in-memory datastore maker.
• The HTTP server now outputs some nice graph (using dagre-d3). Don't
  expect to display very large graphs
• More friendly tag names in the Git backend (no need to prefix
  everything by refs/heads/ anymore)
• Partial support for recursive stores (WIP)

Cleaner external API

• Spring clean-ups in the API. Separation in IrminBranch for
  fork/join operations, IrminSnapshot for snapshot/revert
  operations and IrminDump for import/export operations.
  The later two implementation can be derived automaticaly
  from a base IrminBranch implementation. The update and merge
  operations are supported on each backend
• IrminGit does not depend on unix anymore and can thus be
  compile to javascript or xen with mirage
• No need to have bin_io converter for contents anymore
• No need to have JSON converter for contents anymore
• No more IrminDispatch
• Add an optional branch argument to Irmin.create to use
  an already existing branch
• Fix order of arguments in Irmin.merge

Support for views, speed improvement

• Feature: support for in-memory transactions. They are built
  on top of views.
• Feature: add support for views: these are temporary stores with
  lazy reads + in-memory writes; they can be used to convert back
  and forth an OCaml value into a store, or to have a fast stagging
  area without the need to commit every operation to the store.
• Support custom messages in commit messages
• Improve the IrminMerge API
• Backend: add a 'dispatch' backend for combining multiple backends
  into one. This can be used to have a P2P store where there is
  well-defined mapping between keys and host (as a DHT).
• Fix: limit the number of simulteanous open files in the Git and
  the file-system backend
• Speed-up the in-memory store
• Speed-up the import/export codepath
• Speed-up the reads
• Speed-up IrminValue.Mux
• Deps: use ocaml-sha instead of cryptokit

Support for merge and user-defined contents

• Support for user-defined contents (with custom merge operators)
• Support for merge operations
• Rename IrminTree to IrminNode to reflect the fact that we
  can support arbitrary immutable graphs (it's better if they are
  DAGs but that's not mandatory)
• Rename IrminBlob to IrminContents to reflect the fact that
  we also support structured contents (as JSON objects)
• Support for linking the library without linking to camlp4 as well (#23)

Support for cohttp 0.10.*

0.5.1

Bump version to 0.5.1

Support non-UTF8 blobs and consistent support for watches across backends

• More consistent support for notifications. irmin watch works
  now for all backends.
• Support for different blob formats on the command-line
• Support for JSON blobs
• More flexible irmin fetch command: we can now choose the backend to
  import the data in
• Fix import of Git objects when the blobs were not imported first
• Support non-UTF8 strings as path name and blob contents (for all
  backends, including the JSON one)
• Speed-up the slow tests execution time
• Improve the output graph when objects of different kinds might have
  the same SHA1

API changes and Git backend

• The command-line tool now looks in the environment for the variable
  IRMIN to configure its default backend
• Add a Git backend
• Add Travis CI scripts to the repo
• Use Lwt_bytes and Lwt_unix instead of the custom-made IrminChannel
• Use bin_prot instead of a custom binary protocol
• Major refactoring: Value is now Blob, Revision is now Commit
  and Tag becomes Reference (rational: consistency with Git names)
• Use core_kernel instead of building a custom Identiable.S
• Use dolog instead of a custom log library
• Use mstruct (mutable buffers on top of cstruct) which is now
  released independently