Make SAITS return attention weights in predict()

[WenjieDu]

What does this PR do?

1. making SAITS return attention weights, fixing Visualize self-attention matrix #178;
2. adding deprecation warning and updating docs;

Before submitting

• This PR is made to fix a typo or improve the docs (you can dismiss the other checks if this is the case).
• Was this discussed/approved via a GitHub issue? Please add a link to it if that's the case.
• I have commented my code, particularly in hard-to-understand areas.
• I have written necessary tests and already run them locally.