Gglasso

* Update dependencies, removed old pinned versions for `decorator, sphinx, jinja`

New features:

* Elementwise regularization parameter for single Graphical Lasso. Can be used by passing `lambda1_mask` in `reg_params` (should be a array of nonnegative numbers that has same dimension as covariance matrix `S`).

Enhancements:

* New example in example gallery, showing the usage of Functional Graphical Lasso: https://gglasso.readthedocs.io/en/latest/auto_examples/plot_fsgl_example.html#sphx-glr-auto-examples-plot-fsgl-example-py

Minor Changes:

* remove usage of `np.float` as this is deprecated. Instead use `np.float16,np.float32,...`
* update 3d plot to work with newer versions of `matplotlib` (`gca` was deprecated)
* Update Readthedocs YAML to match the new file requirements (include build section)

Reverts the change on default rescaling behaviour: in `glasso_problem`, by default the solution gets rescaled to covariances. This has the effect that resolving with the regularization parameters found during model selection and with the _unscaled input_, it is not ensured that you obtain the same solution (as the model selection takes place for the _scaled input_).

New features:
- solver for Functional Graphical Lasso
- experimental: thresholding in grid searches (not yet included in `glasso_problem`)

Minor changes:
- changes in defaults for `glasso_problem`
- no rescaling by default
- `store_all` flag allows access to all solutions on the grid

This release is the packages version after review at JOSS.