Xmf uniformisation

Closes "issue" #91 and makes @jregan's life easier when implementing i/o for gparts.

See merge request !89

Tiny fixes to correctly accomodate the switch between SPH variations

Most of it is code re-factoring. I have unified some definitions of the i/o routines and debugging routines.

Solves issues #82 an #83.

See merge request !85

ENGINE_POLICY is #define not variable: parentheses

More haste, less speed: this happens to work with the current Makefile.am, but would break if the order changed. This should fix the Jenkins build.

See merge request !88

MPI affinity: enable, but behave better if not enabled

examples/Makefile.am sets affinity for MPI runs if we can, but examples/main.c doesn't pin the main thread if engine_init won't be pinning the runners. I've checked with top that this means we use multiple cores in both cases, vs. the current scenario of everything ending up on one.

Also restore the getopt -o option lost in a merge, I think.

See merge request !87

* Also restore getopt -o option lost in some merge

Corrected initialisation mistake in MPI mode.

Bug fix. Cells have now their timestep properly initialised and will compute the density of their particles. 

See merge request !86

First version of the multiple time-stepping

Eventually...

That should open up debugging from a larger crowd.

See merge request !79

libnuma, immintrin.h detection in configure script

Add a check for the numa.h header (not just the libnuma binary) and a more robust check for vector.h's inclusion of immintrin.h (currently false negative). Also enable example/main.c's -o option - handled, but absent from getopt call.

This is the last of the non-vectorisation changes from local_position_cache.

See merge request !84