Question about performance

Nobody has claimed this yet.

## Describe the issue

I realise there has already been several issues opened about the performance of `ipympl` (e.g. #55), compared to other libraries that render in jupyter. But I still don't fully understand why the performance is so much worse than using e.g. the `%matplotlib qt` backend.

Take for example the 'Looking glass' example from the [matplotlib docs](https://matplotlib.org/stable/gallery/event_handling/looking_glass.html#sphx-glr-gallery-event-handling-looking-glass-py).
The ineractivity is very nice in the Qt backend (more than enough for our purposes, even if not quite as snappy as javascript based libs such as `bokeh` or `plotly`), but terrible in jupyter lab (laggy and leaving phantom parts of the ellipse around the plot).

Is there really no way to improve the performance of interactivity in `ipympl`? (I have tried to use `fig.canvas.draw_idle()` instead of `.draw()`, and it helps a little but we are still so far behind the Qt backend.

Many thanks.

## Versions

```
3.9.5 (default, Jun 4 2021, 12:28:51)
[GCC 7.5.0]
ipympl version: 0.8.2
Selected Jupyter core packages...
IPython : 7.29.0
ipykernel : 6.5.0
ipywidgets : 7.6.5
jupyter_client : 7.0.6
jupyter_core : 4.9.1
jupyter_server : 1.11.2
jupyterlab : 3.2.3
nbclient : 0.5.8
nbconvert : 6.3.0
nbformat : 5.1.3
notebook : 6.4.5
qtconsole : not installed
traitlets : 5.1.1
Known nbextensions:
config dir: /home/nvaytet/software/miniconda3/etc/jupyter/nbconfig
notebook section
ipycanvas/extension enabled
- Validating: OK
ipyevents/extension enabled
- Validating: OK
jupyter-datawidgets/extension enabled
- Validating: OK
jupyter-matplotlib/extension enabled
- Validating: OK
jupyter-threejs/extension enabled
- Validating: OK
jupyter_bokeh/extension enabled
- Validating: OK
jupyter_dash/main enabled
- Validating: OK
jupyterlab-plotly/extension enabled
- Validating: OK
jupyter-js-widgets/extension enabled
- Validating: OK
JupyterLab v3.2.3
/home/nvaytet/software/miniconda3/share/jupyter/labextensions
ipycanvas v0.12.0 enabled OK
ipyevents v2.0.1 enabled OK
jupyter-matplotlib v0.10.2 enabled OK
jupyterlab-datawidgets v7.0.0 enabled OK
jupyterlab-plotly v5.4.0 enabled OK
jupyter-threejs v2.3.0 enabled OK (python, pythreejs)
@jupyter-widgets/jupyterlab-manager v3.0.1 enabled OK (python, jupyterlab_widgets)
@bokeh/jupyter_bokeh v3.0.2 enabled OK (python, jupyter_bokeh)

Other labextensions (built into JupyterLab)
app dir: /home/nvaytet/software/miniconda3/share/jupyter/lab
jupyterlab-dash v0.4.0 enabled OK

Build recommended, please run `jupyter lab build`:
jupyterlab-dash needs to be included in build

```

Start by reproducing the Looking Glass example in JupyterLab with ipympl and compare its interaction with the Qt backend. Examine the reported draw_idle behavior and the listed JupyterLab, ipympl, and widget versions; done means identifying a concrete performance cause and defining a measurable improvement in interactive rendering.

Written by the indexing model from the issue text.