Keep the caller's numpy when setting up the Core ML backend

backends/apple/coreml/scripts/install_requirements.sh clones coremltools
and installs its reqs/build.pip before building the mlmodel library.
That file pins numpy==2.0.0 for Python 3.9 to 3.12, so running the script
silently downgrades numpy in the caller's environment.

pandas 3.1.0 now requires numpy>=2.0.2, so after the downgrade
`import pandas` fails, and anything that imports the ExecuTorch devtools
(for example examples/apple/coreml/scripts/export.py) breaks with
"Please upgrade numpy to >= 2.0.2". Every Core ML job in CI hits this.

Configuring and building mlmodel only needs numpy (for its include path)
and setuptools (to build the bundled kmeans1d helper). Install just those,
unpinned, so an existing numpy is left alone.

Test plan: in a fresh Python 3.12 venv with numpy 2.4.6 and pandas 3.1.0,
ran the script before and after this change.
- Before: mlmodel builds, numpy drops to 2.0.0, `import pandas` fails.
- After: mlmodel builds, the protobuf sources are copied, numpy stays
  2.4.6, `import pandas` works.

Author: shoumikhin

backends/apple/coreml/scripts/install_requirements.sh

@@ -75,12 +75,14 @@ if [[ "$(uname)" == "Darwin" ]]; then
 fi
 
 echo "${green}ExecuTorch: Installing coremltools dependencies.${reset}"
+# Configuring coremltools only needs numpy and setuptools. Its reqs/build.pip
+# pins an old numpy, which would downgrade numpy in the caller's environment.
 SDKROOT="$HOST_SDKROOT" \
 CC="$HOST_CC" \
 CXX="$HOST_CXX" \
 CFLAGS="$HOST_CFLAGS" \
 CXXFLAGS="$HOST_CXXFLAGS" \
-python -m pip install -r "$COREMLTOOLS_DIR_PATH/reqs/build.pip"
+python -m pip install numpy setuptools
 STATUS=$?
 if [ $STATUS -ne 0 ]; then
     echo "${red}ExecuTorch: Failed to install coremltools dependencies.${reset}"