Try reinitializing context if updateParametersInContext fails

[lilyminium]

This PR addresses the issue in OpenMM where trying to update a simulation without actually changing parameters triggers a pretty esoteric error. This arose for me during Iteration 0 of trying to reproduce a past TIP4P-FB fit, because update_simulation is called even though the FF has not been optimized yet.

Opening as a draft PR for now to see if existing tests trigger the error -- if not, I'll come up with my own test :-)

I ran a timing test and reinitializing the context is about an order of magnitude slower than updateParametersInContext -- reinitializing takes ~35 ms compared to 4.6 ms for updateParametersInContext. So I think the try/except is much more preferable than reinitializing every time.

Code below, and notebook PDF

from openmm.app import *
from openmm import *
from openmm.unit import *
from sys import stdout
import random

pdb = PDBFile('liquid.pdb')
forcefield = ForceField('amber14/tip3pfb.xml')
system = forcefield.createSystem(pdb.topology, nonbondedMethod=PME,
        nonbondedCutoff=0.9*nanometer, constraints=HBonds)
integrator = LangevinMiddleIntegrator(300*kelvin, 1/picosecond, 0.004*picoseconds)
simulation = Simulation(pdb.topology, system, integrator)
simulation.context.setPositions(pdb.positions)

def update_parameters():
    for i in range(simulation.system.getNumForces()):
        if hasattr(simulation.system.getForce(i),'updateParametersInContext'):
            simulation.system.getForce(i).updateParametersInContext(simulation.context)

def reinitialize_context():
    simulation.context.reinitialize(preserveState=True)

def randomize_parameters_and_update(func):
    for fc in simulation.system.getForces():
        if "Nonbond" in type(fc).__name__:
            units = [x.unit for x in fc.getParticleParameters(0)]
            for i in range(fc.getNumParticles()):
                q_ = random.random()
                sigma_ = random.random()
                eps_ = random.random()
                fc.setParticleParameters(i, q_ * units[0], sigma_ * units[1], eps_ * units[2])
    func()

%timeit randomize_parameters_and_update(update_parameters)
%timeit randomize_parameters_and_update(reinitialize_context)

time-reinitialization.pdf

[lilyminium]

I realised that this error may be dependent on CUDA (at least, that's the only way I could trigger the error originally). I did verify that the 9925286 commit triggers that error on a CUDA machine, then verified that 267b442 fixes it. I'm not sure how better to test this on GitHub CI :/