Bump version to 0.1.12 (magenta#657)

Add pianoroll_rnn_nade model. (magenta#640)

* Add rnn-nade model.

* Updates for compatibility

* Limit the batch size based on the beam size and branch factor.

* lint

* Add support for _Layer and Layer depending on availability.

* Add support for _Layer and Layer depending on availability.

* Add README

* Respond to reviewer comments.

* Respond to reviewer comments.

* Rename rnn_nade to pianoroll_rnn_nade for clarity and consistency.

* Rename rnn_nade to pianoroll_rnn_nade for clarity and consistency.

* lint

* Rename rnn_nade to pianoroll_rnn_nade for clarity and consistency.

* Add support for state tuples in bundles.

* Lint

* Lint

* Add missing dep

* Flatten state tuples for collections.

* Simplifications.

* Simplify

* Lint

Prepare for 0.1.10 release (magenta#560)

* Bump version to 0.1.10.

Also switch autoinstaller script to install tensorflow as a part of the
pip depencencies for magenta, rather than downloading a specific .whl
file. Installing via pip is now the recommended installation method.

* Update README.md to install tensorflow via pip.

bump version to 0.1.9 (magenta#504)

bump to 0.1.8 (magenta#469)

Bump version to 0.1.7. (magenta#465)

Also update tensorflow dependency to 0.12.0rc1.

Bump version to 0.1.6 (magenta#370)

increment magenta version to 0.1.5 (magenta#333)

Merge RNN models. (magenta#272)

* Upgrade mido to 1.1.17

* Move encoder/decoders

* Merge melody RNNs.

* Re-add moved files.

* Move melodyencoderdecoders to own file.

* Move melodyencoderdecoders to own file.

* Update sequene generator and magenta_midi to work with merged model.

* Update docs.

* update pip build rules

* Add back 'KeyMelodyEncoderDecoder'

* Add copyright.

* Fix lint errors.

* Fix lint errors.

* Fix visibility.

* Fix internal deps.

* Fix lint errors.

* Bug fixes.

* Fix typo

* don't process bundle file path when None

* Remove bazel from melody_rnn README.

* Push version.

Bump version to 0.1.3 (magenta#270)

Cut a new release prior to announcing the pip package.