Improve configure-time clangd support and shared Doxygen layout

[sean-parent]

Summary

• Generate exact public-header consumer commands during CMake configure using excluded indexing targets, with ownership validation, aliases, generated headers, and nested dependency support.
• Add missing-file VSCode CMake Tools settings for preset-based database copying while preserving existing workspace configuration.
• Add a compact shared Doxygen layout with Reference/Index navigation, description-first groups, inherited default pages, and a documented layout-warning review workflow. Require Doxygen 1.13.1+ for documentation and preserve native Windows documentation exit codes.
• Remove externally managed worktree tasks and tokensave hooks from VSCode configuration, and ignore generated build directories.
• Add Windows/MSVC, Linux/GCC, and macOS/Apple Clang regression coverage.
• Keep regression-harness warning options compatible with both older CMake and CMake 4.4+, which deprecates the legacy unused-CLI flag.

Test plan

Fresh local validation at 0c33bbb8bd596828408cac19527ee3cd9f13cca0 on Windows/MSVC 19.51, CMake 4.3.1, Ninja 1.13.2, Doxygen 1.18.0, clangd 21.1.6, and Python 3.13.14 for protocol checks:

• cmake -P tests/install/CMakeLists.txt: 32 passed, 0 failed; one pre-existing manual-only case remains skipped.
• cmake -P tests/install/test_provider_merge.cmake: all 5 provider-merging checks passed.
• cmake -P tests/install/test_nested_install.cmake: all 11 nested install/package cases passed, including static/shared build/install/downstream consumption.
• cmake -P tests/setup/test_target_type.cmake: all 4 target-type cases passed.
• cmake -P tests/setup/test_runtime_dlls.cmake: custom shared/static executable deployment passed.
• cmake -P tests/setup/test_setup_version_resolution.cmake: scaffold version-resolution regression passed.
• cmake -DCPP_LIBRARY_TEST_CLANGD=ON -DCPP_LIBRARY_TEST_COMPILER_ID=MSVC -P tests/setup/test_clangd.cmake: configure-only exact header commands, fresh-clangd exact definition ranges, consumer flags, generated/nested headers, exclusion/install behavior, settings preservation, and preset switching passed.
• cmake -P tests/docs/test_docs.cmake: all 15 configuration/rendering/review-policy cases passed.
• git diff --check: passed.
• CI at the exact head: all seven jobs passed, including Linux/GCC, macOS/Apple Clang, Windows runner coverage, documentation, and the Ubuntu scaffold/install/find_package integration.

Boundaries

• Live VSCode CMake Tools database copying was not driven through the editor UI; generated settings, preservation, and configure-only preset databases are covered automatically.
• Active multi-configuration generators are rejected instead of choosing ambiguous header commands; single-config Ninja/Makefiles or opt-out are required.
• Local validation used the tool versions above, not the minimum CMake 3.24/Doxygen 1.13.1 versions.
• WARN_LAYOUT_FILE = NO intentionally suppresses layout diagnostics. Layout edits must temporarily enable those warnings and review the complete list, as documented.

Generated with GitHub Copilot CLI

[sean-parent]

Talos local review

Verdict: No actionable issues found in the reviewed changes.

Reviewed commit: 0c33bbb8bd596828408cac19527ee3cd9f13cca0
Scope: Complete 27-file PR diff against main, merge base 2068f2e6522ead4577f147f886e264b66f7db07e. No binary/generated files skipped.
Reviewed by: Security, Correctness, Standards, Architecture, Operations, Documentation, followed by Challenge. All six baseline personas completed; the challenge pass processed the empty candidate set.
Counts: 0 candidate observations; 0 dropped or merged; 0 accepted findings; 0 inline comments posted; 0 overflow or unanchorable findings. No duplicate findings or recurrences to post.

Persona routing: static roster (DCAP routing off; rerun with --dcap to enable).

Project context: Loaded the complete merge-base CLAUDE.md; no rule files skipped, no matching GitHub custom instructions, and no active OpenSpec changes.

Verification: All eight documented local CMake regression scripts passed freshly at this commit on Windows/MSVC, including fresh-clangd definition checks and 15 documentation cases. The dependency-mapping suite reports 32 passed, 0 failed, with its existing manual-only case skipped. Exact-head CI completed successfully: all seven jobs passed, including Windows/MSVC, Linux/GCC, macOS/Apple Clang, documentation, and the scaffold/build/install/find_package round trip. The initial CMake 4.4 deprecation failure was corrected in this reviewed head and the platform jobs reran successfully.

Coverage limitations: Static persona review is not proof of defect absence, and an empty-set challenge does not independently re-review the code. Live VSCode CMake Tools UI copying and the minimum supported CMake/Doxygen versions were not exercised locally. The supplementary Scout packet is incomplete and scope-inconsistent (65 compared files versus the authoritative 27-file diff); reviewers did not rely on its classifications as proof of findings or coverage. Its affected-test list below is not a validated execution plan; validation used this repository's documented scripts and CI jobs.

Scout

Used: risk HIGH, evidence MEDIUM, status INCOMPLETE (scout 0.9.130, <1s).
Cross-file leads given to reviewers: 7 changed symbol(s) with uses outside the diff, 0 new value(s) not handled everywhere, 0 possibly unused symbol(s).

Tests to run (Scout: 45 Must Run test(s), in these files):

• tests/install/fixtures/consumer/examples/example.cpp
  Scout left 44 affected-test entries out of its packet; for the full list run scout affected-tests -r 'D:/repos/github.com/stlab/cpp-library/.claude/worktrees/improve-clangd-support' -s compare -b origin/main.