Make Dask and Distributed optional dependencies

Dask was de facto optional already (see the guarded import in
tsfresh/feature_extraction/data.py), but before this commit it was
installed by default.

Distributed is imported only to support Dask, when the classes
ClusterDaskDistribution or LocalDaskDistribution are used, so it is not
needed if Dask is not imported.

To install tsfresh with those dependencies, one can do:

    pip install -e ".[dask]"

This is needed, for example, to run the test suite. The documentation
has been updated accordingly.

docs/text/how_to_contribute.rst

@@ -47,7 +47,7 @@ Install all the relevant python packages with
 .. code::
 
  cd /path/to/tsfresh
- pip install -e ".[testing]"
+ pip install -e ".[dask,testing]"
  pre-commit install
 
 

docs/text/quick_start.rst

@@ -13,6 +13,15 @@ As the compiled tsfresh package is hosted on the Python Package Index (PyPI) you
 
  pip install tsfresh
 
+If you need to work with large time series data that may not fit in memory, install tsfresh with
+`Dask <https://www.dask.org>`_:
+
+.. code:: shell
+
+ pip install tsfresh[dask]
+
+See also :ref:`large-data-label`.
+
 
 Dive in
 -------

setup.cfg

@@ -42,8 +42,6 @@ install_requires =
  patsy>=0.4.1
  scikit-learn>=0.22.0
  tqdm>=4.10.0
- dask[dataframe]>=2.9.0
- distributed>=2.11.0
  stumpy>=1.7.2
  cloudpickle
 
@@ -56,6 +54,9 @@ exclude =
  tests*
 
 [options.extras_require]
+dask =
+ dask[dataframe]>=2.9.0
+ distributed>=2.11.0
 matrixprofile =
  matrixprofile>=1.1.10,<2.0.0