feat: add full GLSL, HLSL, and shader build support

[Hungbocluaqua]

What changed

Adds first-class shader and shader-build intelligence across the CodeGraph pipeline:

• GLSL and Vulkan GLSL support for standard, ray-tracing, mesh, task, and OpenUSD GLSLFX files
• HLSL support for functions, overloads, semantics, attributes, resources, register/space coordinates, and RTXDI-style include contexts
• pinned ABI-15 tree-sitter GLSL/HLSL grammars with reproducible provenance and third-party notices
• shader preprocessing, macro declarations/references, transitive include resolution, conditional variants, and include-scoped call/resource matching
• shader resources including uniform/storage blocks, push constants, specialization constants, payloads, hit attributes, descriptor coordinates, and Vulkan attributes
• conservative C/C++ to shader file, entry-point, descriptor, specialization, push-constant, and vertex-location relationships
• CMake and PowerShell build-script indexing, including module imports, dot-sourcing, cross-script calls, shader variants, and SPIR-V references
• target-scoped shader callers, impact, affected-test discovery, and explore output with file/line disambiguation
• C++ local lambda extraction and scope recovery needed by Vulkan resource-building code
• documentation, language artwork, configuration guidance, corpus tooling, and regression coverage

Why

Shader-heavy C++ projects currently lose their dependency graph at the CPU/GPU and build-system boundaries. This makes source lookup useful but leaves callers, impact, affected tests, descriptor ownership, compilation targets, and application-specific shader bridge implementations incomplete or ambiguous.

The implementation was developed against a Vulkan ray-tracing engine and NVIDIA RTXDI, with repeated regression testing for overloads, conditional macros, include contexts, application bridges, shader entry points, PowerShell modules, CMake variants, and C++ pipeline ownership.

References the language roadmap in #648 and the original GLSL request in #685. The closed #688 PR was used only as a regression reference; this implementation is based directly on current main.

Validation

• npm run build
• package dry run: 750 packaged files; both GLSL/HLSL WASM grammars and THIRD_PARTY_NOTICES.md present
• focused explore/disambiguation sweep: 81/81 tests passed
• changed-area sweep: 650/651 tests passed; the sole failure was a Windows EBUSY cleanup of a temporary uninitialized-submodule directory
• post-fix dynamic/shader sweep: 58/58 tests passed
• Vulkan corpus: 227 shader-family files parsed twice deterministically, expected 99 first-party GLSL files, zero extraction failures
• RTXDI corpus: 282 shader-family files parsed twice deterministically, zero extraction failures
• real-index probes confirm exact main -> trace_path and main -> RTXDI_DITemporalResampling Flow, Relationships, caller source, and callee source

The full Windows pre-push suite also exposed existing Windows file-lock/global-hook-path failures in temporary MCP, worktree, SQLite, ArkTS, and git-hook tests. One genuine dynamic-dispatch labeling regression found in that run was fixed in the final commit and covered by the 58-test post-fix sweep. Docker Desktop's Linux daemon was unavailable locally, so this draft relies on GitHub CI for the authoritative clean Linux full-suite result.

Review notes

This is intentionally a draft because the implementation spans language extraction, resolution, graph traversal, retrieval, and build integration. The commits are organized as implementation/tests, documentation/evaluation assets, and a final dynamic-dispatch precision fix. I am happy to split the draft into a staged PR series if that is easier to review.