LTR improvements

[afoucret]

This PR contains two improvements for the LTR feature in Eland.

• Feature logger now extract NaN when a query does not match a query. It is consistent with what is done in Elasticsearch.

• A specific method (import_ltr_model) has been added to import LTR model with support for a ltr_model_config object of class LTRModelConfig. The inference_config params is removed from the import_model method

Example:

ed.ml.MLModel.import_ltr_model(
  es_client=es_client,
  model=ranker,
  model_id='ltr-model-xgb',
  ltr_model_config=ltr_model_config,
  es_if_exists = 'replace'
)

The previous version was more difficult to use (and could lead to inconsistencies):

ed.ml.MLModel.import_model(
  es_client=es_client,
  model=ranker,
  model_id='ltr-model-xgb',
  feature_names=ltr_model_config.feature_names,
  inference_config=ltr_model_config.to_dict(),
  es_if_exists = 'replace'
)

[pquentin]

buildkite test this please