fix working directory alteration

[hoangnv735]

Fix working directory alteration when initializing VnCoreNLP.

When creating a VnCoreNLP instance, the working directory is changed to save_dir. This behavior cause errors when using with some package like Flask or Hydra.
After loading the annotation model, I added a line to change it back to self.current_working_dir to fix.

I have tried some workaround to avoid changing the working directory, but I finally realized that the Java code of VnCoreNLP load models by related path. So I guess working directory alteration is inevitable now.
Below is the Java code from VnCoreNLP

public class WordSegmenter {
    private  Node root;
    private static WordSegmenter wordSegmenter = null;
    public final static Logger LOGGER = Logger.getLogger(WordSegmenter.class);
    public WordSegmenter()
            throws IOException {
        LOGGER.info("Loading Word Segmentation model");
        String modelPath = System.getProperty("user.dir") + "/models/wordsegmenter/wordsegmenter.rdr";
        if (!new File(modelPath).exists())
            throw new IOException("WordSegmenter: " + modelPath + " is not found!");

        this.constructTreeFromRulesFile(modelPath);
    }