Change marktasks to rebuild

In order to obtain a correct cell->h_max, we need to rebuild at the beginning.

Fix #552 (closed)

@matthieu: By the way after 100 steps the largest relative errors are 5 (single particle) and about 1 (large number of particle). I think it is just due to the chaotic behavior of our simulations and I am not really worried about it.

This is all good. Let's just wait for another MR to go in first as it is being tested right now.

mentioned in commit a0f68fd8

merged