Do not use -march=native in CI

[DanShaders]

GitHub-provided CI runners are heterogeneous, so artifacts compiled with -march=native on a newer runner may fail to run on an older runner if ccache reuses same object files. To combat this, we introduce (and use in CI) TON_USE_CI_BASELINE_CPU CMake flag that overrides -march=native with some (relatively old) baseline CPU arch.

[github-actions]

Findings

• CMakeLists.txt:83-93 – TON_USE_CI_BASELINE_CPU now unconditionally adds -mcpu=apple-m1 on macOS/arm64. Apple Clang <13 rejects that flag (previous code gated the M1 flag on clang≥13), so enabling the new option on older toolchains will break the build. Consider keeping the version guard or falling back to a more widely supported arch for older compilers.
• CMakeLists.txt:95 – TON_ARCH is injected via add_compile_options(-march=${TON_ARCH}) without the prior CHECK_CXX_COMPILER_FLAG guard. A mistyped/unsupported arch (or an empty value) now fails during compilation with a less clear error instead of a configure-time check. Suggest restoring validation to give immediate, actionable feedback.

Notes

• x86_64 baseline uses -march=x86-64-v3; if anyone enables the flag with a compiler too old to support that level, it will fail at compile time.

[DanShaders]

This requires passing -march=${TON_ARCH} carefully to all dependencies as well, which is extraordinarily annoying.