Make Dask an optional depedency

[thehomebrewnerd]

PR #248 adds dask[dataframe] as a dependency. This will only be needed for users that want to use Woodwork with Dask DataFrames. We should move this to an optional dependency.

As Featuretools and EvalML will use Woodwork, the solution we implement here should also work with those libraries. Will also need to consider how dependencies will be handled on conda-forge - do we want separate versions there, or just one that includes all dependencies?

For Reference:

• https://evalml.alteryx.com/en/latest/install.html#Pip-with-all-dependencies
• https://featuretools.alteryx.com/en/stable/install.html

[gsheni]

After discussing this internally, we want pip install woodwork[dask] to install dask dependencies to use Dask & Woodwork.
We can use extra_requires, like Featuretools:

• https://github.com/FeatureLabs/featuretools/blob/main/setup.py#L9