Fix checkpoints

[RemiLehe]

This is a reimplementation of PR #162, decided after off-line discussion with @MKirchen.

From the original PR:

The checkpoint / restart feature was broken:

• In load_species the weights were multiplied by the charge q. This should not be done anymore.
• If a diag_period was at the same time as a checkpoint_period, a wrong diagnostic was written at the first step (i_step == 0) after the restart (wrong charge/currents). With this PR, a check is performed based on a new attribute (is_restarted) of the Simulation object.

The current PR fixes the second issue by swapping the order of the diagnostics and the particle exchange (particle exchange first, then diagnostics). Since the particle exchange phase is always followed by a deposition of charge (including immediately after a restart, see the remark on i_step==0), the charge will now always be available for the diagnostics. In addition, I added a similar line for the currents (only done when i_step==0, not for each particle exchange).

As a side-effect, note that this also takes care of the fact that the charge/currents were not deposited on the very first diagnostics (even without restart).