MPI fixes

[conradtchan]

Type of PR:
Bug fix, modifications to existing code

Description:
This PR includes several bugfixes. 3 fixes have been bundled from other PRs to reduce the testing load once the phantom benchmarks suite has been fixed.

#239:
Bug introduced in #96 , particles are assigned new IDs if a file without ID is read. IDs are erroneously duplicated across MPI tasks. This is fixed by using an offset to ensure all IDs are unique.

#240:
npartoftypetopt=reduceall_mpi('+',npartoftype) is used in many places in the code, but it is always calculated on the fly. It only needs to be calculated after the number of any particle type on a task has changed. This PR stores it as a global variable in part, and defines a subroutine to recalculate it from npartoftype.

#241:
The summary variables wake, floor, and tolv were not reduced across MPI tasks, resulting in incorrect printout. This did not seem to have any effect on the actual results.

Other small bugfixes:

• replace MPI_TYPE_STRUCT with MPI_TYPE_CREATE_STRUCT: MPI_TYPE_STRUCT was deprecated since MPI 2.0 and replaced with MPI_TYPE_CREATE_STRUCT. It has exactly the same interface and does exactly the same thing. And then MPI_TYPE_STRUCT was removed from the standard in MPI 3.0 But MPI_TYPE_STRUCT still works, but gives the incorrect result
• reduce dtextforce across mpi tasks, and only print to log from the master task
• remove unused variable declaration (dtype_old)
• reduce box size across MPI tasks in initial
• use ntot instead of npart for nskip, since npart is only the local particle count
• use ntot instead of npart to print dtlog
• change particle memory allocation ratio, and make consistent between fortran and hdf5 outputs
• optimisations to step leapfrog: don't reduce ptmass variables unless necessary
• use kind=8 for np in timestep routines

Testing:
Full test suite

Did you run the bots? yes, pre-commit