Fix various issues with MPI and partitioning

These are two bugs and one feature in the current handling of repartitioning
and MPI tasks.

Extracted from current developments so they do not get lost.

See merge request !436

Fix int-in-bool-context with gcc-7.2

With GCC-7.2.0, I am getting the following error:
<code>
scheduler.c:1606:14: error: ‘<<’ in boolean context, did you mean ‘<’ ? [-Werror=int-in-bool-context]
     if (!((1 << t->type)) || t->skip) continue;
          ~~~~^~~~~~~~~~~~
</code>

See merge request !444

Large parallel hdf5

Fix to #72. 

This is a workaround to the limitation of parallel-HDF5. The low-level MPI-IO implementations limit writes to 2GB per rank (irrespective of the total amount being written across all nodes). 

The solution involves writing chunks of 2GB (or in practice 2'000'000'000Bytes) and then repeat for the remaining chunks, if any, by shifting the position to write of each node in the file and in memory by 2GB. Ranks that did not pass the threshold just write nothing.
In realistic scenarios we won't need more than a handful of iterations.

See merge request !438

Added SWIFT code group

Adds ```Code/Code = "SWIFT"``` in the output files which can be used as a check to see if the output was generated by SWIFT.



See merge request !439

Added SWIFT code group to the output files to enable a quick check from other programs if the output was generated by SWIFT

 Only skip tasks with zero costs, note previous method had a bug as task_type_sub_pair was always skipped

Been like this for quite some time, the effect would seem to be we have no been using any edge weights

(cherry picked from commit 8458c3c3)

These are meant to be larger than other tasks, keep recv at a slighly lower cost as that is a speculative call (to MPI_Test), whereas sends are expecting a result

(cherry picked from commit bfcebf24)

Also send dx_max_part and dx_max_gpart alongside ti_end_min at the end of a time-step when running with MPI

Implements the solution to #367. Instead of just sending the time-step we send the distances moved by any part and gpart in that cell. 

I have also done some cleaning up around the packing and unpacking and improved the documentation.

See merge request !434

Correct vec neighbour find

Fixes neighbour find bug in the vectorised version of the density interactions.

See merge request !435

Added an extra test case to testActivePair.sh to include the RNG seed that creates a vectorised neighbour find bug.

Also send dx_max_part and dx_max_gpart alongside ti_end_min at the end of a time-step when running with MPI.

Fixes to the gravity code

This fixes #363. The long-range gravity task has to consider all the cells that were assigned to it at the rebuild time. 

Additional improvements:
 - Drift the multipoles while activating the tasks,
 - Avoid FPEs in the FFT task,
 - Adjust the number of tasks per top-level required to run.

See merge request !430

Small cosmetic improvements to the documentation of the FMM code.



See merge request !432

Added debugging checks to verify that the particles are in the correct frame after the shift in DOPAIR1 and DOPAIR2. The check takes into account directions and drift since last rebuild.

Corrected the same check in the vectorized versions of the pair interactions. This was not using the correct criterion for the maximal displacement any particle can have had since the last rebuild.

See merge request !431

Added debugging checks to verify that the particles are in the correct frame after the shift in DOPAIR1 and DOPAIR2. The check takes into account directions and drift since last rebuild.