Non-bonded interactions set_params() is broken

[jngrad]

The non-bonded interactions set_params() method has undocumented behavior that can lead to unphysical potentials.

Once a non-bonded interaction is activated with valid parameters using set_params(), the second call to set_params() silently changes behavior: required parameters become optional, and missing arguments are populated using the values currently in the core, completely bypassing the default_params() method.

This unexpected behavior can compromise simulation scripts in two ways:

1. The user intends to reset an interaction with only the required parameters, expecting the default parameters to populate the missing non-required parameters: the old parameters are silently recycled
2. The user discovers this undocumented behavior and intends to use it to update one parameter of a LennardJonesInteraction or GenericLennardJonesInteraction potential that was previously set up with shift="auto": the shift parameter is not automatically recalculated and the potential has a jump in energy.

Here is a MWE to reproduce the bug in the second scenario:

import espressomd
import numpy as np
import matplotlib.pyplot as plt

system = espressomd.System(box_l=3 * [10.])
system.time_step = .1

def sample():
    start_pos = [0., 0., 0.]
    step = np.array([0.01 / 600, 0., 0.])
    system.part.clear()
    p0, p1 = system.part.add(pos=[start_pos, start_pos + 600 * step], type=[0, 0])
    xval = []
    yval = []
    for _ in np.arange(180):
        p1.pos = p1.pos + step
        xval.append(np.linalg.norm(p1.pos - p0.pos))
        yval.append(system.analysis.energy()["non_bonded"])
    return np.array(xval), np.array(yval)

system.non_bonded_inter[0, 0].lennard_jones.set_params(epsilon=1.92, sigma=0.05, cutoff=0.11, shift="auto")
# trigger bug: forget to use shift="auto", so the shift isn't recalculated for the new cutoff
system.non_bonded_inter[0, 0].lennard_jones.set_params(cutoff=0.012)
xval1, yval1 = sample()
# fix bug: manually add shift="auto"
system.non_bonded_inter[0, 0].lennard_jones.set_params(cutoff=0.012, shift="auto")
xval2, yval2 = sample()

plt.plot(xval2, yval2, label="set_params(cutoff=0.012, shift=\"auto\")")
plt.plot(xval1, yval1, label="set_params(cutoff=0.012)")
plt.xlabel("Inter-particle distance")
plt.ylabel("Energy")
plt.legend()
plt.show()