Use the threadpool to parallelize operations in the particle-splitting code

Implements #641 (closed).

All the loops over the global particle arrays have been parallelized. I have also added alignment information to help the compiler use faster memcpy implementations.

Ran this and haven't uncovered any issues, but I'm not seeing any significant speed up with respect to master. I've even gone as far as lowering the mass limit so that we get a lot more splits, and still nothing, which is odd. Did you see a speed up?

I did not really expect one. :) In most cases, we only do the first loop and find that there are no particles to split anyway.

Think I'm happy with this and will accept as it is no worse and has the potential to be better. Any last thoughts from you?

No, I am also happy. Thanks!

merged

mentioned in commit c4d7c5e0

Good, accepted. I'm on holiday for the rest of the week, so no more pushes for a while.