[BUG] current_potential_energy() is stale after run() unless energy_frequency is set

[lohedges]

Dynamics.current_potential_energy() returns the energy from before the last run() block unless that block happened to save energies. The integrator advances positions without going through SOMMContext.setPositions(), so the context's _energy_cache is never invalidated, and the only clear on the dynamics path is gated behind if save_energy:.

import sire as sr, openmm

mols = sr.load_test_files("ala.top", "ala.crd")
d = mols.dynamics(timestep="1fs", temper

def direct(d):
    return (
        d.context()
        .getState(getEnergy=True)
        .getPotentialEnergy()
        .value_in_unit(openmm.unit.kilocalorie_per_mole)
    )

print(d.current_potential_energy().value
d.run("5ps")
print(d.current_potential_energy().value
d.run("5ps", energy_frequency="1ps")
print(d.current_potential_energy().value
d.run("5ps")
print(d.current_potential_energy().value

                                       cached         direct
initial                            -5987.8102     -5987.8102
after run(5ps)                     -5987.8102     -6194.3530   <-- stale
after run(5ps, energy_frequency)   -6094.2036     -6094.2036   <-- fine
after run(5ps) again               -6094.2036     -6132.6996   <-- stale