Add ability to return a predictor instead of modifying in place.

Currently, to load a predictor after optimization, you need to do it in two lines:
```python
predictor = dspy.Predict("question -> answer")
predictor.load(saved_path)
```

Desired behavior
```python
predictor = dspy.Predict("question -> answer").load(saved_path, return_self=True)
```

This fix is as adding a bool to the `load` method of `BaseModule` to return self.