create an extra_require group for soft dependencies

[mattijn]

The following quote is originally posted by @jonmmease in #2816 (comment):

We could add vl-convert-python to an extra_require group and then update the docs to install using that group. For example, we could add vl-convert-python, altair-viewer, and vega-datasets to an extra_require group named full and then instruct people in the docs to install altair with

pip install "altair[full]"

[mattijn]

Include pyarrow as a soft dependency, #2888

[binste]

Consider a pyodide specific dependency set, see #3118 (comment)

[joelostblom]

I think also including vegafusion in the extra_require group would be helpful. And creating a conda package for altair-full would also be welcome (although if there is a better way that does not require maintaining a separate meta package then that would be great).

I was reminded about this today when a colleague asks me how to include altair in their env file and I sent them six lines of what to install depending on what they are doing, which is not ideal:

dependencies:
    - altair=5.1.2
    - vl-convert-python  # For saving altair charts as static images
    - vegafusion  # For working with charts > 5,000 graphical objects
    - vegafusion-python-embed  # Same as the previous one
    - vegafusion-jupyter  # For working with charts > 100,000 graphical objects
    - vega_datasets  # Example data

[binste]

I like the idea of the dependency groups! See #3309 (comment)

Regarding the naming of the dependency groups, I like the approach of pandas where they use all instead of full to refer to installing all dependency groups. Feels more natural to me.