Add Word Embedding (with Pooling) to benchmark

[dcecchini]

Add to the embedding benchmark the word embedding models as well. To do that, we use a pooling strategy to transform the word embeddings into a sentence embedding. Common pooling strategies are:

• Averaging each word embedding from the sentence
• Using the special tag [CLS] or <s> (depends on the model) as a representation of the sentence (CLS pooling)