New experiment: multi-LJ grid

[fxcoudert]

The idea is simple: if you're interested in multiple LJ probes, then they differ by two things:

• value of epsilon, which is a multiplicative constant in the total energy
• value of sigma, which is more complicated

But we can produce voxel grids for multiple values of sigma, in a range (sigma_min, sigma_max, num_sigma) when we produce a grid. This will increase the grid size, but it will not increase the grid calculation time much.

Time to generate a grid with:

• 1 sigma value (old code): 1.87 s
• 10 sigma values: 2.01 s
• 20 sigma values: 2.05 s
• 50 sigma values (extreme case): 2.43 s

The result is a multi-channel grid, with size ngrid x ngrid x ngrid x num_sigma.