[feature request] add JSON to export and import the model

[maximskorik]

Description

I am working on adding the Spec2Vec package to the data processing pipeline @RECETOX. The implementation would include passing a trained model to Spec2Vec to compute the similarity. Due to the security vulnerability of the Pickle data format, we decided to implement a custom JSON exporter and importer for the model inside our pipeline. But we thought you might be interested in adding this functionality to the package (e.g., for #77).

Advantages: no security vulnerabilities – it will be safe to compute scores based on publicly available models, etc.
Disadvantages: such a model will not be trainable anymore and can only be used to compute scores

Implementation

• Create a lightweight subclass of gensim.model.Word2Vec, that will only store the data required by Spec2Vec to compute the scores.
• Implement exporting and importing for this subclass. The model will be stored in two files: a .json file for metadata, and .npy or .npz for storing the model's embeddings as numpy.ndarray or scipy.sparse respectively.