CI: run the Windows tests through ctest with --repeat until-pass (#1205)

The Windows and pixi-Windows jobs ran the gtest executable once, so a single
wall-clock timing test missing its budget on a loaded runner failed the whole
PR (RepeatTestAsync.RepeatTestAsync, SequenceTripleActionTest.TripleAction).
Use ctest --repeat until-pass:3, as the macOS job already does.

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

Author: facontidavide

.github/workflows/cmake_windows.yml

@@ -49,4 +49,8 @@ jobs:
 
     - name: Run tests
       working-directory: ${{ github.workspace }}/build
-      run: $env:PATH+=";${{ matrix.build_type }}"; tests/${{ matrix.build_type }}/behaviortree_cpp_test.exe
+      # Some tests assert tight wall-clock timing, which the shared runners can
+      # not always meet: --repeat until-pass absorbs that flakiness (as on macOS).
+      run: |
+        $env:PATH += ";${{ github.workspace }}\build\${{ matrix.build_type }}"
+        ctest -C ${{ matrix.build_type }} --output-on-failure --repeat until-pass:3

pixi.toml

@@ -7,7 +7,7 @@ channels = ["conda-forge"]
 platforms = ["linux-64", "win-64"]
 
 [target.win-64.tasks]
-test = "PATH=\"$PATH;build/Release\" build/tests/Release/behaviortree_cpp_test.exe"
+test = "PATH=\"$PATH;$PIXI_PROJECT_ROOT/build/Release\" ctest --test-dir build -C Release --output-on-failure --repeat until-pass:3"
 
 [target.linux-64.tasks]
 test = "./build/tests/behaviortree_cpp_test"