Disable CEA_BUILD_TESTING on Windows bindc CI legs

[djkees]

Summary

The three Windows bindc matrix legs (windows-latest / intel / intel, windows-latest / gcc / gcc, windows-latest / msvc / intel) configure with -DCEA_BUILD_TESTING=ON but never actually built GFE/pFUnit, so cea_core_test was silently absent there. That gap was invisible until the "fail visibly when pFUnit is missing" change (#209, now on main) added a placeholder test that fails loudly instead — which is exactly what surfaced this as a Windows CI failure on main right after #209 merged.

An earlier version of this PR tried building GFE/pFUnit on Windows to get real coverage there, but that uncovered a chain of Windows-only gaps in GFE's own build (skipped GFE checkout, gfortran's missing _WIN32 predefine, a missing CMAKE_PREFIX_PATH, and finally a missing m4 dependency for the gFTL submodule) with no guarantee that was the last one, especially for the msvc/intel leg's unusual mixed toolchain. Given the bindc job's actual purpose is exercising C-binding builds across toolchains, not the Fortran core suite, this PR now just makes the Windows legs stop claiming test coverage they never had.

Changes

• Configure (Ninja): pass -DCEA_BUILD_TESTING=OFF instead of ON when runner.os == 'Windows' (Linux/macOS unaffected).
• Configure (msvc): pass -DCEA_BUILD_TESTING=OFF (this step only ever runs on Windows).

Testing

• Verified via this PR's own CI run across all three Windows legs.

Compatibility / Numerical behavior

• No expected changes to numerical results
• Expected changes (explain and provide validation)

CI-workflow-only change; no solver, algorithm, or data file touched. No test coverage is actually lost — cea_core_test was never built or run on these Windows legs before either; it was just silently absent instead of openly reporting so. The Fortran core suite remains covered by the pfunit-tests job (ubuntu-22.04 / gcc 11).

---

Drafted with Claude's assistance.

• Root cause was verified against the actual GitHub Actions logs for run 33775466923, the follow-up run on main (33775501053), and this PR's own two CI runs (which caught the additional GFE/pFUnit gaps before landing on this simpler fix).
• Confirmed via source/CMakeLists.txt and the top-level CMakeLists.txt that cea_core_test is only ever registered when CEA_BUILD_TESTING AND PFUNIT_FOUND, so turning testing off where pFUnit was never built removes only the loud placeholder failure, not any real test.