kmeans weights variance parallelization

See merge request !54

[gmm] Simply calls to functions by passing GMM machine around

Closes #38

See merge request !52

[gmm] deduplicate EM training code

Closes #39

See merge request !51

improve variance estimation speed in kmeans, convert data to proper arrays

See merge request !50

Fix np memory issues and rename k_means to kmeans

Closes #42, #40, and #37

See merge request !49

[kmeans] refactor e-m to have simpler graphs for dask

Closes #36

See merge request !48

Fixes to kmeans and gmm

See merge request !47

Allow setting the k-means oversampling_factor.
Naming of GMM map parameters.
GMM.load() testing.

Improvements to the new API and docs

Closes #35

See merge request !46

- Added pre-commit checks
- Dropped old dependencies
- Brought all classes and functions to bob.learn.em
- Make KMeans.centroids_ available as KMeans.means
- Implement a load method for GMMMachine

Support h5py v3

Closes #33

See merge request !45