Vectorise kernel

Should fix issue #154. Includes updates to `testKernel.c` that test `kernel_deval_vec` against the serial `kernel_deval`.
I have tested it with Cubic Spline, Wendland C2 + C6 kernels.

See merge request !150

Made the kernel test part of the testing suite and checks that the vector/scalar values are consistent.

Put all the hydro task dependencies linking into a single function.

This implements @tt 's suggestion to get all the hydro-task dependency construction in one single place. 

See merge request !152

Fixed missing dependency between init and density in the case of sub-tasks.

Not sure whether these dependencies disappeared at some stage or whether they have never been there but they are clearly missing.

That fixes #153 and potentially the MPI issues if they are due to missing links. Can't really test it from here. 
@pdraper that might be of interest to you.  

See merge request !151

Recover : for y command-line parameter and update task plotting scripts for external gravity

Fixes a problem from the parameter files branch and updates the
plotting scripts for external gravity.

See merge request !149

Initialise the entropy only once.

Exactly what is says on the tin. 


See merge request !148

Gravity particles

Ok, that's another big one...  This includes all the changes by @tt and @jregan regarding the use of an external potential.

The main changes are:
 - Update to the `unit` with some more const statements.
 - Addition of a set of physical constants and of an object to convert them in the chosen system of units.
 - A new external gravity task and its dependencies.
 - This task calls functions in a new file `potential.h` where users can define their external potentials.
 - A new test case that just makes a bunch of particles orbit a point mass.

See merge request !143

Pasc paper

Please let me know as soon as possible if these changes sufficiently cover the two comments, and if you are OK with the wording.

Your changes from `pasc_paper_final` (!146) have been merged into this branch because I suck at git.

See merge request !147

x_old is now a x_diff since the last tree build. It is also implemented for the gparts.

Two changes here following the discussion in #122:

 - `x_old` is replaced by `x_diff`, i.e. the offset since the last tree-build. That makes the drift task a bit strange as they have to move `x` in one direction and `x_diff` in the other.
 - The `gpart` structure also gains an `x_old` field. This is necessary for the cases where we only have gparts and no parts. If we don't do this, the tree reconstruction is never triggered in `engine_marktask()`.

A neat consequence is that we can store `x_diff` as float and not as double. So the `xpart` structure goes down to 32bytes in size (after alignment).

See merge request !145