[WIP] Add method for differentiable 1D histogram

[RemiLehe]

This adds a method to compute 1D histograms for ParticleBeam

TODO:

• Discuss the interface/implementation:
  - should it be a method of ParticleBeam only? => yes, but open an issue to implement for both beams
  - what normalization to use by default? => Copy interface of numpy.historgram
  - should we reuse this function in plot_1d_distribution (with `method='histogram') => yes
  - check vectorization
  - check that it runs on GPU
• Add an automated test

Here is a quick code snippet on how to use/test this on GPU:

import cheetah
import torch
import matplotlib.pyplot as plt

device = "mps"

# Initialize a particle beam
incoming = cheetah.ParticleBeam.from_parameters(
    num_particles=100_000,
    sigma_p = torch.tensor(1e-2, requires_grad=True),
    energy = torch.tensor(1e9)
).to(device)

# Check that both histogram methods give the same result
h1 = incoming.get_1d_histogram(
    dimension="p",
    bins=100,
    bin_range=(-3e-2, 3e-2),
    method="histogram",
    )

h2 = incoming.get_1d_histogram(
    dimension="p",
    bins=100,
    bin_range=(-3e-2, 3e-2),
    method="kde",
    )

plt.plot(h1.detach().cpu().numpy())
plt.plot(h2.detach().cpu().numpy())

# Check that the histogram obtained with KDE is differentiable
h2.sum().backward()

[RemiLehe]

This can be used in plot_1d_distribution