Documentation updates for 1.0.0-beta4

[AlexDBlack]

Here's a summary list of features and changes in beta4 that the docs will need to be updated for:

ND4J

• ND4J DataTypes support/changes
• Nd4j. new methods: createFromArray, DataType overloads for existing methods
• Update memory management page (periodic GC disabled by default; INDArray.close() method)
• Check pages for rank 0/1 support/changes
• Review, update and consolidate old pages
• Numpy NPZ support
• Reduction axis change (in op not as exec arg)

DL4J

• DL4J: MKL-DNN support
• Notes on DataTypes (datatype config) and datatype conversion.
• Update quickref guide: new layers (attention etc). L1/L2/Weight decay - classes now. Weight init: classes now. GELU activation. BERT iterator. Embedding layer word vectors based initialization. Capsule network layers. Cifar10DataSetIterator. Wassrstein loss.
• Evaluation: update for rank 4 support, and axis argument
• Check docs on ParallelInference; add notes on ModelAdapter
• NLP pages: add references/links for BERT WordPiece tokenizers. Notes on next format serialization
• UI page - add section on multi-user/multi-session support
• Update CUDA page for new CUDA versions

DataVec

• PythonTransform
• Add new transforms: FirstDigitTransform, TokenizerBagOfWordsTermSequenceIndexTransform, GazeteerTransform
• StreamInputSplit

SameDiff

• SameDiff docs in general
  • Tensorflow import - examples, what is/isn't supported, etc
• TensorFlowImportValidator, GraphTransformUtil, convertToVariable/convertToConstant
• SameDiff UI

Arbiter

• Genetic algorithm search