Fast 2D DBSCAN

Fast C++ implementation of DBSCAN for 2D data that follows the sklearn interface.

It's not exactly fast yet. TODO:

• Add tests
• Add benchmark (maybe together with tests using Catch2)
• Use 2D grid-based hashing instead of kd-tree for neighborhood lookup
• Use SIMD vectorization
• port to GPU
• build and export Python wheels
• support CMake for building