Mistake in the `rshift` for the `cj` side of the stellar loops

It looks like there might be a mistake in the star pair loop in the case where we have a periodic BC.

On this line https://gitlab.cosma.dur.ac.uk/swift/swiftsim/blob/master/src/runner_doiact_functions_stars.h#L456 we end up deducting rshift twice (once explicitly and once via dx_max_rshift). That is wrong. It also affects the loop exit condition on line 456.

@chaikin spotted this and proposed to change the sign of rhsift on line 432. This would cancel all the rshift on the LHS of line 456 and only keep the one that is correctly used in di_max on the RHS.

What do you think?