No clip to 0-1

[shinyoung-yi]

Generally images may not belong to the range [0, 1] for several reasons:

• HDR images (captured real image or rendered synthetic image)
• Difference map of two images
• Signed values for differentiable rendering
• Other non-RGB radiance quantities can be considered as images including normal vectors, wave optics-based images, etc.

[haruishi43]

@class255 thank you for your work. I merged your revisions, but I thought clipping the values based on the input tensor/ndarray seemed a better approach than using the output. Sometimes, the values will overflow or become weird based on the interpolation methods, especially around the image borders. I also added clip_output argument for all functions based on a previous pull request, so if you pass clip_output=False to any of the functions, it will skip the clipping entirely. Hope this helps

[shinyoung-yi]

Thanks for accepting this PR.
Now I realized that clipping by np.nanmin(out) and np.nanmax(out) does not have any meaning and using horizon in the current version is much better.
Thank you again for your work!