Histogram argument not inherited when using global dictionary parameters

Description of the issue

When using a global dictionary for key-worded arguments in corner.corner(), the method will not take into account new parameters (for color for instance) if it already exists in a global instance passed as a parameter to hist_kwargs.

This is because hist_kwargs is written on the parameter passed, not on a copied instance. Therefore, some of its argument may already exist if it was modified during a previous corner.corner call, typically in a loop.

Steps to reproduce

Here is a minimal code reproducing to issue:

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

fig = plt.figure()

corner_kwargs = dict(
    fig=fig,
    show_titles=True,
    plot_contours=True,
    fill_contours=False,
    plot_density=False,
    plot_datapoints=False,
    title_fmt=".2f",
    hist_kwargs={"density": True}, # HERE
    label_kwargs={"fontsize": 18},
    linewidths = 1.5,
    alpha=0.5
)
colors = ["r", "g", "b"]
ndim, nsamples = 2, 10000
for color in colors:
    samples = np.random.randn(ndim * nsamples).reshape([nsamples, ndim])
    corner.corner(samples, color=color,
                  # hist_kwargs={"density": True}, # HERE
                  **corner_kwargs)

Expected behavior

One would expect the color to change at each iteration for the histogram, but it doesn't (although correlation plots are not affected).

Workaround

To bypass this issue, one can define a new dictionary instance to pass at each iteration, such as in the commented line #HERE above.

However this prevents from creating clean common parameter dictionary when working with a large number of common arguments.

Start at the corner.corner() entry point and trace how hist_kwargs is handled across repeated calls using the reproduction example. Ensure the shared dictionary is not modified between calls, then rerun the loop to verify that each histogram uses the current color while correlation plots remain unaffected.