Use one-less cores in preset workflows (pytorch#11073)

### Summary
must go faster


![image](https://github.com/user-attachments/assets/218b6500-6276-4c34-95ca-3cf52c8a6f2b)


### Test plan
CI

cc @larryliu0820

Author: jathu

.github/workflows/build-presets.yml

@@ -27,12 +27,12 @@ jobs:
       runner: macos-latest-xlarge
       python-version: 3.12
       submodules: recursive
-      timeout: 900
+      timeout: 90
       script: |
         set -eux
         ${CONDA_RUN} ./install_requirements.sh > /dev/null
         ${CONDA_RUN} cmake --preset ${{ matrix.preset }}
-        ${CONDA_RUN} cmake --build cmake-out -j16
+        ${CONDA_RUN} cmake --build cmake-out -j$(( $(sysctl -n hw.ncpu) - 1 ))
 
   linux:
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
@@ -65,4 +65,4 @@ jobs:
 
         ./install_requirements.sh > /dev/null
         cmake --preset ${{ matrix.preset }}
-        cmake --build cmake-out --parallel
+        cmake --build cmake-out -j$(( $(nproc) - 1 ))