Fix gpart id/pointer

Making a union of the gpart ID and the part pointer seemed like a good idea at first, but it's just an accident waiting to happen, as the pointers are not guaranteed to be less than 2^63. I therefore suggest the following:

• Store the gpart ID as a ptrdiff_t.
• If the gpart ID is positive, it's just the particle's ID.
• If the gpart ID is negative, then it is the negative offset in the space->parts[] array of its part counterpart.

This storage format has the additional advantage that it makes it easier to re-link the particles if we ever do have to send them around between nodes, e.g. by making the offsets relative to whatever array the data is stored/sent in.

This is currently just a proposal. Please let me know what you think and if there is no opposition I will change this as soon as the initial gpart branch has been merged (!116 (merged)).