build(release): package MIInfer v0.2.0 baseline

[GionaGranchelli]

Summary

• Isolate research benchmarks/tools from standard Release packaging.
• Add the documented CPack TGZ install path and package/first-run smoke checks.
• Preserve V2-0047 qualification evidence.

Validation

• V2-0047 PASS evidence is committed in experiments/V2-0047-release-packaging.md.
• Branch head is 919dd64c9603ab539d81b285928cfde3ca3a81f2.
• Diff contains packaging, documentation, graph output, and release smoke-test changes; no inference-performance source changes.
• Existing untracked gpucore.3330886 is intentionally excluded.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The non-trivial CMake EXCLUDE_FROM_ALL build-graph logic determines exactly what ships in the release package, so a human should confirm release contents before approval despite no defects being found.

Review effort: Balanced
Findings: None

What changed in this PR

This PR packages the MIInfer v0.2.0 release by separating research/benchmark build targets from the standard release build, curating the installed file set, and hardening the package and first-run smoke tests. It supports the project's packaging milestone without changing any inference, kernel, numerical, or public CLI behavior.

Changes:

• Adds MIINFER_BUILD_BENCHMARKS/MIINFER_BUILD_RESEARCH_TOOLS/MIINFER_BUILD_TESTS options and excludes bench/tools/non-release-test executables from the default all target, so a broken GCC‑16 benchmark no longer blocks cmake --target package; also curates the install manifest to two binaries, installer, license, README, and three docs.
• Strengthens scripts/test-package.sh (exact file-manifest allowlist, --version identity checks, default-prefix + repeat-install verification, host-only CTest gate that skips before the archive exists) and scripts/test-cli-first-run.sh (/healthz + /v1/models checks, .gguf-stripped model_id, unauthenticated non-loopback rejection check).
• Adds docs/release.md, a V2-0047 experiment record, and updates README.md/developer-guide.md/graph output to match the new release/docs layout.

File	Description
CMakeLists.txt	New build-graph options and EXCLUDE_FROM_ALL logic to isolate research targets; curated install list; registers package-archive CTest gate.
scripts/​test-package.sh	Adds manifest allowlist, version/identity checks, default + repeat install verification, diagnostics-retaining cleanup, and SKIP-before-built behavior.
scripts/​test-cli-first-run.sh	Adds /healthz and /v1/models assertions, stem-based model_id, and an unauthenticated LAN-serving rejection check.
docs/​release.md	New install/build-from-source instructions and qualified V2‑0045 performance scope table.
README.md	Updates install steps (extract + run installer) and points to the new release/hardware/CLI docs.
developer-guide.md	Documents the new CMake build-graph options.
experiments/​V2-0047-release-packaging.md	Records the packaging qualification evidence and verification transcript.
graphify-out/​manifest.json, graphify-out/​GRAPH_REPORT.md	Regenerated knowledge-graph artifacts reflecting the doc/build changes.

I verified the key correctness points: the test-package.sh allowlist matches the curated install(...) destinations (share/doc/MIInfer/...), the version parse matches CPACK_PACKAGE_FILE_NAME for the gfx906/Linux target, the CMake exclusion regex relies on relative source paths (which all add_executable calls use) with correct exemptions for miinfer/miinfer-device-info/miinfer-m12-gdn-oracle, the model_id stem matches the server's std::filesystem::path(model_path).stem(), the /healthz//readyz response strings match the assertions, and the docs/release.md numbers match the V2-0045 qualification record. I found no concrete, objective defects to comment on.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.