Incorrect features for MusicCNN models (pypi version)

[romi1502]

The pypi version of musicnn still contains a bug that was fixed
in this commit, but not deployed on pypi.
Then extracting the penultimate layer output using the extractor function with extract_features=True results in incorrect outputs with the pypi version.

[suncerock]

Similar issue found in both the pypi version and the main branch
Think this line

musicnn/musicnn/extractor.py

Line 216 in 516acb2

predicted_tags, timbral_, temporal_, midend1_, midend2_, midend3_, mean_pool_, max_pool_, penultimate_ = tf_out

should be
predicted_tags, timbral_, temporal_, cnn1_, cnn2_, cnn3_, mean_pool_, max_pool_, penultimate_ = tf_out