Laplacian, cosine distance, auto-tune, and constrained spectral clustering

v0.2.0 is a complete overhaul of the SpectralCluster library.

List of changes and new features:

• Use 2-space indent Google internal coding style.
• Use options classes instead of long list of arguments.
• Support different types of Laplacian matrices.
• Support different types of distances for K-Means, such as cosine distance.
• Support auto-tune of the p-percentile.
• Implement constrained spectral clustering.
• Use Enum wherever possible.
• Add configs.py as reference to configurations used in our papers.
• Add permutation invariance for tests.