Fix CI build: accommodate CMake 4.x and skip upstream chai REPL

[leftibot]

Summary

Fixes the CI failures from #34 (tracked in #33). The build breakage is rooted in the outdated CMakeLists.txt that ships with ChaiScript 6.1.0 and gets pulled in via FetchContent.

Two independent CMake issues:

• macOS Configure error — runners now ship CMake 4.x, which removes support for cmake_minimum_required(VERSION < 3.5). ChaiScript 6.1.0's root CMakeLists.txt still declares VERSION 2.8 and configure aborts. Fixed by setting CMAKE_POLICY_VERSION_MINIMUM 3.5 just before add_subdirectory, forcing an acceptable floor for the transitive project.

• Windows MSVC Build error — ChaiScript 6.1.0's src/main.cpp (the chai REPL) has a missing <chrono> include and fails to compile under current MSVC ('high_resolution_clock': is not a class or namespace name). The REPL target is declared unconditionally in the subproject, so EXCLUDE_FROM_ALL on add_subdirectory keeps it out of the default build. chaiscript_extras only consumes ChaiScript's headers, so no functionality is lost.

Drive-by modernization: bump cmake_minimum_required to 3.16, explicitly opt into CMP0169 OLD so the intentional FetchContent_Populate call stays quiet on CMake 3.30+, and drop the now-unused BUILD_MODULES ON override (modules aren't needed by the tests, and EXCLUDE_FROM_ALL would skip them regardless).

Out of scope

The Linux ASAN/UBSAN job still fails on string_methods_test with a heap-use-after-free originating in chaiscript::extras::string_methods::split when accessed via a chained index expression (e.g. split(",")[1]). That is a runtime bug in the extras+ChaiScript 6.1.0 interaction — not a CMake issue — and should be tracked separately.

Test plan

• Local clean configure on CMake 3.31 — no errors, deprecation warnings silenced
• cmake --build (Debug) — succeeds, chai REPL not built
• cmake --build (Release) — succeeds, chai REPL not built
• ctest (Debug) — both tests pass
• ctest (Release) — both tests pass
• GitHub Actions CI — macOS/Windows/Linux non-sanitizer jobs should go green; Linux sanitizers will still fail on the pre-existing runtime bug noted above

🤖 Generated with Claude Code