Separate computation and update steps into two distinct functions of the module's external API

[giacomofiorin]

Discussed in #746

^{Originally posted by HanatoK October 30, 2024}
I think the separation is required for using Monte-Carlo barostat. In the Monte-Carlo barostat, we don't increase the step number, and evaluate the forces multiple times at the same step. PLUMED has two entrance functions for the MD engine, namely (i) calculate() and (ii) update(). Both of them are supposed to be called in a forward manner, but the former should only calculate the CVs, biasing energy and forces, while the latter is supposed to update the internal state of the biases. In the OpenMM-PLUMED interface, the two routines are called by the following code (https://github.com/openmm/openmm-plumed/blob/a3e935bc88ed8771f17f6f4d60588b81789b9c89/openmmapi/src/PlumedForceImpl.cpp#L143-L148):

    plumed_cmd(plumedmain, "performCalcNoUpdate", NULL);
    if (step != lastStepIndex) {
        plumed_cmd(plumedmain, "update", NULL);
        lastStepIndex = step;
    }

As you can see, update() will be only called after the step counter changed.

Now that NAMD is going to support the Monte-Carlo barostat in the GPU-resident code path. I wish Colvars can be compatible with it (or can be used in OpenMM in the future), so I think it is important to separate the calculation of CVs and biases and the update of their internal states.