Why is 'nointerpolation' applied to 'double' types and default interpolation set for FS inputs but not VS outputs?

**Question/Issue**

While reading the source code in [glslang/HLSL/hlslParseHelper.cpp](https://github.com/KhronosGroup/glslang/blob/b58249a96fce25cce5d440b265b58fead80d1377/glslang/HLSL/hlslParseHelper.cpp#L2327), I noticed that variables of type `double` are marked with the `nointerpolation` qualifier. Additionally, when handling fragment shader (FS) inputs, the code assigns a default interpolation qualifier by default, but for vertex shader (VS) outputs, this default interpolation is not set.

**Questions:**
1. Why is the `nointerpolation` qualifier applied to variables of type `double`?
2. Why is a default interpolation assigned to fragment shader inputs, but not to vertex shader outputs, even though this seems inconsistent?

**Context:**
- I am seeking clarification on the design or technical motivation behind these behaviors in the codebase.
- Is this decision based on HLSL/GLSL specification requirements, or are there other reasons?

Any insight into the reasoning or relevant specification references would be appreciated.

Thank you!

Start at glslang/HLSL/hlslParseHelper.cpp around line 2327 and trace the handling of double variables and interpolation qualifiers for fragment inputs and vertex outputs. Consult the relevant HLSL and GLSL specification sections, then document the rationale and any intentional asymmetry with precise references.

Written by the indexing model from the issue text.