Add note to docs about embeddings changing after tensorflow 2 / kapre + librosa upgrade

[beasteers]

@magdalenafuentes made a good point that we should add a note to the openl3 docs that mention that the embeddings are different after 0.4.0 so that people aren't caught off guard. And maybe we can include the plot that @hohsiangwu made showing the performance comparison between the different frontends and tensorflow versions