Km3pipe

~~~~~~~~~~~~~~~~~~~
* ``TimeslicePump`` now supports the readout of any stream ("L0", "L1", "SN"...)
* Minor bugfixes (km3pipe has no attribute named hardware...)

~~~~~~~~~~~~~~~~~~~
* add loguniform distribution (``kp.math``)
* add contextmanager for pumps (``with HDF5Pump(fname) as h5: print(h5[0])``)
* clean up makefile / installer docs
* debug compilation/makefile issues

~~~~~~~~~~~~~~~~~~~
* Windows compatible version of `sys.peak_memory`. KM3Pipe should now compile
and work under windows...
* fix issues with hit indexing when merging multiple h5 files

~~~~~~~~~~~~~~~~~~~
* ``core.pyx`` and ``tools.pyx`` have been "depyxed"
* ``Calibration.apply**`` (should) always returns the hits
* ``Module.finish`` (and thus the pipeline!) actually return something now!
* ``Calibration`` shortcut removed from ``km3pipe``, so now you have to use
``from km3pipe.calib import Calibration`` or just ``kp.calib.Calibration``
if you ``importe km3pipe as kp``.
This change was needed to be able to import __km3pipe__ in Julia.
* ``kp.io.hdf5.HDF5Pump`` now accepts the path of a boolean cut mask,
e.g. ``cut_mask='/pid/survives_precut'``. If the bool mask is false, the
event is skipped.

~~~~~~~~~~~~~~~~~~~
* add a makefile
* flake8 all the things
* make compatible for upcoming numpy 1.14
* add ``nb2shpx`` util for notebook -> sphinx gallery exampe
* some pandas bits and bobs

~~~~~~~~~~~~~~~~~~~
* Improved CLB raw data readout
* Pipelines now return a ``finish blob`` which contains the return values
of each modules finish method (this is for Tommaso)
* ``TimesliceParser`` now reads all timeslice streams (L0, L1, L2, SN)
* ``TimesliceParser`` now returns the blob even if it was not able to parse
the data
* ``TMCHRepump`` now has an iterator interface
* Fixed bug in ``StreamDS`` where it tried to create a session in Lyon and
failed. Now it uses the permanent session which was created by Cristiano
* Some smaller bugfixes and name-consistency-changes