Clamp PBR surface opacity before shader construction

[tdavidovicNV]

The PBR specification defines surface cutout opacity (O) over the range [0, 1], but does not specify how values outside that range are handled.

Consider this shader:

<?xml version="1.0"?>
<materialx version="1.39">
  <oren_nayar_diffuse_bsdf name="diffuse" type="BSDF" />

  <surface name="background" type="surfaceshader">
    <input name="bsdf" type="BSDF" nodename="diffuse" />
    <input name="opacity" type="float" value="0.2" />
  </surface>

  <surface name="foreground" type="surfaceshader">
    <input name="bsdf" type="BSDF" nodename="diffuse" />
    <input name="opacity" type="float" value="2.0" />
  </surface>

  <mix name="surface_mix" type="surfaceshader">
    <input name="bg" type="surfaceshader" nodename="background" />
    <input name="fg" type="surfaceshader" nodename="foreground" />
    <input name="mix" type="float" value="0.25" />
  </mix>

  <surfacematerial name="material" type="material">
    <input
      name="surfaceshader"
      type="surfaceshader"
      nodename="surface_mix"
    />
  </surfacematerial>
</materialx>

OSL clamps opacity at each leaf surface, giving:

mix(clamp(0.2, 0, 1), clamp(2.0, 0, 1), 0.25) = 0.4

GLSL and MDL effectively clamp only the final result, giving:

clamp(mix(0.2, 2.0, 0.25), 0, 1) = 0.65

Clamping at each surface node is preferable because opacity represents coverage in [0, 1]. A surface with opacity 2.0 has no meaningful coverage interpretation, and allowing that invalid value to participate in later surface operations can produce valid-looking but unintuitive downstream results.

This PR:

• Defines the cutout opacity as O = clamp(opacity, 0, 1) before the surface shader is constructed.
• Adds matching [0, 1] UI bounds and documentation to ND_surface.
• Applies the clamp in the common hardware implementation used by GLSL, MSL, and Slang.
• Applies the same behavior to the MDL implementation, aligning it with the existing OSL behavior.

If clamping is not the intended policy, the specification should instead explicitly declare out-of-range opacity inputs undefined. In either case, the current disagreement between backends should be resolved or documented.