Add hierarchical clustering and dendrograms to Heatmap

[VinzentRisch]

Add optional hierarchical clustering for heatmap rows and columns, including coordinated row and column dendrograms.

• SciPy can handle the clustering calculation, but this introduces a SciPy dependency.
• All distance metrics and linkage methods that are exposed in SciPy can be included as options for the parameters distance and method parameters. Default distance = euclidian, method = complete.
• Parameter "clustering" with options: features, samples, both, none

High-level workflow:

input matrix
  → optionally standardize values
  → calculate row and column pairwise distances with scipy.spatial.distance.pdist
  → compute hierarchical linkage with scipy.cluster.hierarchy.linkage
  → obtain dendrogram geometry and leaf order with scipy.cluster.hierarchy.dendrogram(no_plot=True)
  → reorder heatmap rows and columns by leaf order
  → convert SciPy dendrogram coordinates into line-segment data
  → render the heatmap and dendrogram line segments in Vega

The Vega specification can use generic rule marks to render every dendrogram branch from the generated segment data; branches do not need to be manually hard-coded.

Codex code Example:
scipy-vega-clustered-heatmap.py

Plot:

[lizgehret]

Notes for myself:

• scipy won't be an issue to add to vizard dependency-wise bc types already depends on it (& vizard depends on types)
• @ebolyen suggested a singular heatmap that can optionally take clustering; if included, the dendrograms will be drawn (else the regular heatmap will be rendered). This will add some complexity to the existing heatmap to accommodate this, but I think this would likely be the preferred user experience.