Previous-step total forces are calculated using current-step inverse gradients

[giacomofiorin]

To keep parallel performance high, total atomic forces cannot be communicated to Colvars until all calculations of a MD step come in. For this reason, total atomic forces are always those from the previous time step, and the total colllective force at step 0 is not available.

The total atomic forces are used to compute total collective forces (labeled as "ft" in the colvars.traj output) by using the inverse atomic gradients (IAGs). However, the IAGs are typically calculated using atomic gradients from the current time step. Such inconsistency creates a small error, which has affected total force calculation and ABF starting from early versions of the Colvars module. The same is true for the Jacobian term, which uses the current value of the collective variable.

A solution to this can be to compute the IAGs and the Jacobian term, store them in the atom_group container (see also #61), and use them at the beginning of the next MD step.

[jhenin]

Presumed fixed by 833c580.