Change the Model Serving documentation to document the new GPU selection user interface elements.

You can now use the UI to specify:

• The GPU request (nvidia.com/gpu: "1") an integer between 0 and 8
• The GPU limit (nvidia.com/gpu: "1") an integer between 0 and 8
• The GPU toleration (to tolerate a taint. default taint: "nvidia.com/gpu")

If/When I edit any of these values, I'd expect the YAML definitions to be updated in such a way as to trigger a restart of the matching pods.