text embedding average gives more weight to the last batch

[StefanBogdan]

The current implementation gives more weight for the last batch if it has less sentences than all the previous ones. This can be fixed by adding all the sentence embeddings and normalize it by the number of sentences.