Merge branch 'smoothing_length_derivative' into 'master'

Improved multi-timestep SPH

This brings a range of improvement, mostly minor and of code clarity nature to the code. 
It solves:

- Issue #84. Particles are now <128 Bytes in size.
- Issue #85. Smoothing-lengths now evolve more accurately and we implement a time-step criterion more conservative than default Gadget-2 but that is part of EAGLE.
- Issue #88. We now have a 'minimal' hydro implementation that can be used both as a template and an example. The doxygen documentation is based on that version. The documentation is now very extensive. This model is a poor SPH implementation and should only be used to understand the structure of the code and not be seriously run.
- Issue #95. The engine_collect_kick() function now correctly skips empty cells.
- Issue #96. Conserved quantities are now written to a file. We need gravity before this really get useful.
- The description of the SPH flavour in the HDF5 files is now moved to the hydro_io.h files for consistency.


Most of it are straight-forward changes. Note that this does not fix the issue we have in MPI mode.

See merge request !90