Select a native HLSL validator for artifacts requiring 16-bit scalar types

[NripeshN]

Problem

Project and regression validation can fall back from DXC to glslangValidator for generated HLSL. That fallback is not valid when the artifact requires native DirectX 16-bit scalar types such as int16_t or uint16_t.

For example, the Metal nested-reference regression emits an HLSL helper with int16_t parameters. DXC accepts this profile with -enable-16bit-types, but glslang rejects the declaration during HLSL parsing even when invoked with --hlsl-enable-16bit-types.

This also applies to repository-porting workloads whose reports declare directx.native-16bit-types, including MLX kernels at commit 4367c73b60541ddd5a266ce4644fd93d20223b6e.

Required behavior

• Detect HLSL artifacts that require the native 16-bit DirectX profile.
• Select DXC with -enable-16bit-types for those artifacts.
• When DXC is unavailable, report a structured validator-unavailable result instead of treating glslang as an equivalent validator.
• Keep glslang fallback validation for HLSL profiles it can parse faithfully.
• Cover direct toolchain APIs, project validation, and regression-test helpers with tests.

Acceptance criteria

• Native-16-bit HLSL is never submitted to an incompatible glslang fallback.
• Reports distinguish tool unavailability from shader compilation failure.
• Existing non-native HLSL glslang validation remains covered.
• A reduced int16_t helper fixture passes with DXC and produces the expected unavailable result without DXC.