Dynamic memory reallocation for MPI

[conradtchan]

Currently, when the particle array allocation is exhausted, the code halts. #214 solves this problem by increasing the fixed allocation size, but this is in general an inefficient use of memory.

It's worth considering whether the arrays should be automatically expanded when the limit is reached. On HPC clusters, this may not be very useful because the resource allocation provided by the scheduler is fixed, so attempting to allocate beyond this will result in an OOM kill. It may just be easier to allocate the full amount of memory requested from the scheduler.