Explainerdashboard

Breaking Changes
- Now always have to pass a specific port when terminating a JupyterDash-based
(i.e. inline, external or jupyterlab) dashboard: `ExplainerDashboard.terminate(port=8050)`
- but now also works as a classmethod, so don't have to instantiate an
actual dashboard just to terminate one!
- ExplainerComponent `_register_components` has been renamed to `component_callbacks`
to avoid the confusing underscore

New Features
- new: `ClassifierPredictionSummaryComponent`,`RegressionPredictionSummaryComponent`
- already integrated into the individual predictions tab
- also added a piechart with predictions
- Wrapped all the ExplainerComponents in `dbc.Card` for a cleaner look to the dashboard.
- added subtitles to all components

Improvements
- using `go.Scattergl` instead of `go.Scatter` for some plots which should improve
performance with larger datasets
- `ExplainerDashboard.terminate()` is now a classmethod, so don't have to build
an ExplainerDashboard instance in order to terminate a running JupyterDash
dashboard.
- added `no_permutations` boolean argument to `ImportancesComponent` (that
you can also pass to `ExplainerDashboard` `**kwargs`)
-


Other Changes
- Added warning that kwargs get passed down the ExplainerComponents
- Added exception when trying to use `ClassifierRandomIndexComponent` with a
`RegressionExplainer` or `RegressionRandomIndexComponent` with a `ClassifierExplainer`
- dashboard now uses Composites directly instead of the ExplainerTabs

Bug Fix:
- When `X `had a `RangeIndex`, the `plot against feature` component would not work. This is now fixed (by casting it to a str index)

Improvement
- hiding orientation dropdown on ShapContributionsGraphComponent by default. (undo by passing `hide_orientation=False`.)

Breaking Changes
- removed `metrics_markdown()` method. Added `metrics_descriptions()` that
describes the metric in words.
- removed `PredsVsColComponent`, `ResidualsVsColComponent` and `ActualVsColComponent`,
these three are now subsumed in `RegressionVsColComponent`.

New Features
- Added tooltips everywhere throughout the dashboard to explainer
components, plots, dropdowns and toggles of the dashboard itself.


Improvements
- changed colors on contributions graph and xgboost trees graph: up=green, down=red
- added `higher_is_better` parameter. Set to False to switch green and red colors.
- Clarified wording on index selector components
- hiding `group cats` toggle everywhere when no cats are passed
- passing `**kwargs` of ExplainerDashbaord down to all all tabs and (sub) components
so that you can configure components from an ExplainerDashboard param.
e.g. `ExplainerDashboard(explainer, higher_is_better=False).run()` will
pass the higher_is_better param down to all components. In the case of the
ShapContributionsGraphComponent and the XGBoostDecisionTrees component
this will cause the red and green colors to flip (normally green is up
and red is down.)

New Features
- added (very limited) sklearn.Pipeline support. You can pass a Pipeline as
`model` parameter as long as the pipeline either:
1. Does not add, remove or reorders any input columns
2. has a .get_feature_names() method that returns the new column names
(this is currently beings debated in sklearn SLEP007)
- added cutoff slider to CumulativePrecisionComponent
- For RegressionExplainer added ActualVsColComponent and PredsVsColComponent
in order to investigate partial correlations between y/preds and
various features.
- added `index_name` parameter: name of the index column (defaults to `X.index.name`
or `idxs.name`). So when you pass `index_name="Passenger"`, you get
a "Random Passenger" button on the index selector instead of "Random Index",
etc.

Bug Fixes
- Fixed a number of bugs for when no labels are passed (`y=None`):
- fixing explainer.random_index() for when y is missing
- Hiding label/y/residuals selector in RandomIndexSelectors
- Hiding y/residuals in prediction summary
- Hiding model_summary tab
- Removing permutation importances from dashboard


Improvements
- Seperated labels for "observed" and "average prediction" better in tree plot
- Renamed "actual" to "observed" in prediction summary
- added unique column check for whatif-component with clearer error message
- model metrics now formatted in a nice table
- removed most of the loading spinners as most graphs are not long loads anyway.

New Features
- Explainer parameter `cats` now takes dicts as well as a list where you can specify
your own groups of onehotencoded columns.
- e.g. instead of passing `cats=['Sex']` to group `['Sex_female', 'Sex_male', 'Sex_nan']`
you can now do this explicitly: `cats={'Gender'=['Sex_female', 'Sex_male', 'Sex_nan']}`
- Or combine the two:
`cats=[{'Gender'=['Sex_female', 'Sex_male', 'Sex_nan']}, 'Deck', 'Embarked']`