Km3pipe

~~~~~~~~~~~~~~~~~~
* ``Table`` can now be instantiated with ``fillna=True`` when created from
``dict`` and ``dtype`` where keys in the ``dict`` are missing. Those will
be filled with NaNs.
* The ``Module.only_if`` parameter now also accepts a list of keys, which has
to be present in the blob, otherwise the ``process`` method is not called.
* The ``HDF5Sink`` now also accepts "chunksize", "complib" and "complevel as arguments."

~~~~~~~~~~~~~~~~~~
* Fixed read-in of split tables when shuffling in ``HDF5Pump``

~~~~~~~~~~~~~~~~~~
* A new standard parameter called ``blob_keys=['list', 'of', 'blob', 'keys']``
can now be used to filter the blob keys before passing it to a module
during the cycle

~~~~~~~~~~~~~~~~~~
* The ``RandomState`` is dead, long live the ``GlobalRandomState``!
(We renamed it...)

~~~~~~~~~~~~~~~~~~
* Minor changes in Dockerfile and docs

~~~~~~~~~~~~~~~~~~
* ``io.pandas`` has been removed
* DETX v3 supported (including the ability to
``kp.hardware.Detector.add_comment()`` which are preserved when writing
* DUSJ readout fixed, now every parameter is written by default (with NaNs
if missing)
* ``HDF5Sink`` now only writes instances of ``Table`` and ``NDArray`` to
simplify the implementation and avoid future bugs
* ``HDF5Sink`` now can shuffle the blobs when ``shuffle=True``, additionally
a ``shuffle_function`` can be defined to have more control (mutating).
* ``km3modules.mc.RandomState`` can be used to set the global random seed
of numpy to be able to create reproducible pipelines
* In ``HDF5Pump`` when reading multiple files, each file is only opened when
needed to avoid unneeded memory and computational overhead