Use g++ instead of clang and fix PythonInstall

clang is tested everywhere else, and it's not as routinely up-to-date on
my Jenkins machine.

PythonInstall failed on Linux machines where the build/install
directories had paths with spaces in them.

devtools/ci/jenkins/install.sh

@@ -8,15 +8,14 @@ swig -version
 echo "Using cmake (`which cmake`) version":
 cmake --version
 
-echo "Using clang (`which clang`) version:"
-clang --version
+echo "Using g++ (`which g++`) version:"
+g++ --version
 
-module load cuda/${CUDA_VERSION} conda/jenkins
+module load cuda/9.0
 
 # Constants
 CONDAENV=openmm-test-3.5
-INSTALL_DIRECTORY="${WORKSPACE}/openmm-install"
-SRC_DIRECTORY="${WORKSPACE}/openmm-src" # set in the Jenkins configuration
+INSTALL_DIRECTORY="`pwd`/install"
 export OPENMM_CUDA_COMPILER=`which nvcc`
 
 # Create a conda environment, but clean up after one first. If it doesn't exist, don't complain.
@@ -31,9 +30,8 @@ conda install -yn ${CONDAENV} numpy scipy pytest --quiet
 source activate ${CONDAENV} # enter our new environment
 
 # Build OpenMM
-cd "${SRC_DIRECTORY}"
-cmake -DCMAKE_INSTALL_PREFIX="${INSTALL_DIRECTORY}" -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang .
-make -j4 install
+cmake -DCMAKE_INSTALL_PREFIX="${INSTALL_DIRECTORY}" -DCMAKE_CXX_COMPILER=g++ -DCMAKE_C_COMPILER=gcc .
+make -j6 install
 make PythonInstall
 
 # Now run the tests

wrappers/python/pysetup.cmake.in

 set(ENV{OPENMM_INCLUDE_PATH} "@WRAPPER_INCLUDE_DIRS@")
-file(TO_NATIVE_PATH "@CMAKE_INSTALL_PREFIX@/lib" OPENMM_LIB_PATH)
-set(ENV{OPENMM_LIB_PATH} "${OPENMM_LIB_PATH}")
+set(ENV{OPENMM_LIB_PATH} "@CMAKE_INSTALL_PREFIX@/lib")
 message("OPENMM_LIB_PATH = " $ENV{OPENMM_LIB_PATH})
 message("OPENMM_INCLUDE_PATH = " $ENV{OPENMM_INCLUDE_PATH})
 execute_process(