docs(site): publish llama.cpp v0.2.0 downloads

[LauraGPT]

Summary

• publish the nine verified runtime-llamacpp-v0.2.0 archives in the bilingual deployment guide
• render platform, architecture, backend, direct download, and SHA-256 in a responsive matrix
• replace the source-build-first example with a digest-checked Linux x64 release path
• document the unresolved Windows AMD Vulkan hardware confirmation boundary
• validate optional download records before the site can build

User impact

Users can select and verify the correct CPU, Vulkan, or CUDA package without searching GitHub Releases or guessing which archive matches their machine. The page keeps the AMD Vulkan claim scoped to what has actually been verified.

Validation

• python -m pytest web-pages/product-site/tests -q (68 passed)
• deterministic double build and validate.py passed
• Playwright full suite passed; focused mobile and desktop download-matrix checks passed (2 passed)
• all 9 declared files and SHA-256 values exactly match the public GitHub Release API
• all 9 unauthenticated public asset URLs responded successfully
• signed commit with DCO; git diff --check passed