Dypac

This release includes:
* major bug fixes.
* main class in `dypac` is now named `Dypac`.
* working batch system to limit memory / computing load.
* weighting of grey matter for improved detection of RSNs. Check parameters `grey_matter` and `std_grey_matter`.
* embedding tools to transform to and from the parcellation space. Check new attributes `transform`, `inverse_transform`, `compress` and `score`.
* new demo `dypac_demo.ipynb`
* many bug fixes.

Contributors:
Hanad Shamarke, Désirée Lussier have contributed bug reports and optimization.
François Paugam contributed the initial version of the embedding tools.
Pierre Bellec implemented bug fixes, batching, grey matter weights and embeddings.

There are now two modules:
* `bascpp` features the core "barebone", generic methods.
* `dypac` has the object class for fMRI data

The method has also been changed: now stab_maps are extracted and trimmed within batches, and then matched and averaged across batches by a separate consensus clustering. Hopefully this approach will scale very well.

Important change in the API: `n_states` now controls the total number of states, which before it was the number of states per cluster.

The main method `dypac` is functional, and the notebook `dypac_fmri_compression.ipynb` gives a demo of the method.