Bug fix: make remap deterministic for degenerate input ranges

[bhouston]

Found via https://material-fidelity.ben3d.ca. remap is not deterministic for metal, glsl or osl with them producing different results:

Define deterministic behavior for float remap when the input range is degenerate (inhigh == inlow).

• In GLSL codegen (IM_remap_float_genglsl), add a guard:
  • if abs(inhigh - inlow) <= 1e-8, return outlow
  • else use the standard remap expression
• In OSL support code (mx_remap(float, ...)), add the same guard before division.

Why

remap with a zero-width input range is undefined under the raw formula due to division by zero. Different backends were producing different results for nodes/remap_degenerate because of NaN/Inf handling differences.

This change establishes a stable, cross-backend default:

• degenerate input range -> outlow

Scope

• Limited to float remap behavior in genglsl and genosl.
• Non-degenerate remap behavior is unchanged.

Validation

Re-rendered nodes/remap_degenerate with MaterialX renderers and they all match:

[ld-kerley]

It's worth noting that in OSL "divide-by-zero" is not degenerate. It's defined to return 0.

[jstone-lucasfilm]

See #2914 (comment) for recommendations on resolving this category of undefined behaviors through new specification proposals, so that our per-language implementations have an unambiguous reference to match.