Parameters clipping in *Gradient-based optimization* tutorial prevents optimisation convergence to correct colour.

[KelSolaar]

In https://github.com/mitsuba-renderer/mitsuba-tutorials/blob/master/inverse_rendering/gradient_based_opt.ipynb

    # Post-process the optimized parameters to ensure legal color values.
    opt[key] = dr.clip(opt[key], 0.0, 1.0)

The clipping of the optimized parameters prevents convergence to correct colour:

original_color_r 0.5700680017471313
original_color_g 0.043013498187065125
original_color_b 0.04437059909105301

With clipping:

color/optimisation/r 0.5676486492156982
color/optimisation/g 0
color/optimisation/b 0

Without clipping

color/optimisation/r 0.5676485896110535
color/optimisation/g 0.04339996352791786
color/optimisation/b 0.041679296642541885

Even though the optimiser makes negative excursions, it recovers nicely and converges to the expected result.