Adds the model_type field to TrainedModelConfig for distinguishing between models
that can be loaded via the model loading service and those that require a native process.

This adds a temporary API for doing inference against
a trained model deployment.

Introduces code for re-assembling the individual chunks a model is stored in and 
streaming those chunks to the inference process. Re-uses the 
TrainedModelDefinitionDoc format already defined for boosted tree models

Binary data is stored in lucene base64 encoded, the same data stored in a 
Java string uses 2 bytes (UTF16) to represent each base64 character 
consuming twice the amount of memory required. The compressed
binary representation of the models can stored in ByteReferences
more efficiently. For BWC a new field mapping binary_definition 
is added .ml-inference-* and the index version incremented.

This adds a location field to TrainedModelConfig for large models that cannot be
PUT inline with the config. Large models are reassembled from their location.

Adds tokenisation for BERT models via the WordPiece algorithm using the
vocabulary that defined with the model and introduces the concept of NLP
tasks. Each task is configured with a BERT model supporting that task, 
pre-processing and post-processing is defined by the task. 

Named Entity Recognition and Fill Mask are the 2 task types supported by
this PR