Using ipympl logic inside python shell or ipython?

I have a class that inherits from the `ipywidgets.widgets.VBox` and adds a `figure.canvas` instance of a matplotlib figure to its children like so:

```python
import matplotlib.pyplot as plt
from ipywidgets import widgets

plt.ioff()

class MyVis(widgets.VBox):
def __init__(self):
super().__init__()
self.fig = plt.figure()
self.children = [self.fig.canvas]

m = MyVis()
```

This works fine when I use this class within jupyter notebooks using the `%matplotlib widget` magic before executing the above code.
When executed within ipython I get
`TraitError: Element of the 'children' trait of a MyVis instance must be a Widget, but a value of was specified.`
since of course `ipympl` and its magic are not active.

Since I also want to have unit-tests for this class (and others that make use of this class) using `pytest`, it would be nice to get around this problem.
Can I enable the `ipympl` way of handling figures somehow in the python shell (similar to using the `%matplotlib widget` magic in the notebook) that would allow my tests to run?

Start with the MyVis example and reproduce the TraitError in IPython, then compare it with the notebook setup using the `%matplotlib widget` magic. Check how the matplotlib canvas is selected and determine what test setup would let the pytest cases use a compatible widget canvas outside Jupyter; done means the class can be instantiated and tested without the Tk canvas error.

Written by the indexing model from the issue text.