Reducing the number of time-step communications

Only active the ti_end comms for part and gpart if we are not running with limiters. When limiters are on, we do a second round of exchanges anyway.

This likely has a rather minimal effect but can help if we think the problem is that we swamped the MPI library. The better fix I am looking at next is to dramatically reduce the number of comms we do during the second round of exchanges.