Various fixes

- Do not enable CPU device together with OpenCL/CUDA when device name is not
  specified.
- Adopt output log parser fr the changed path tracing message.
- Solve issue with using same short name of command line argument for single
  device specification.
- Reshuffle include statements so it's kind of possible to use bundle python.
  Still need to work this properly, but that's for later.

benchmark/foundation/system_info.py

@@ -1,6 +1,5 @@
 import json
 import platform
-import psutil
 import subprocess
 import sys
 
@@ -77,6 +76,7 @@ def getNumPhysicalCPUs():
 
 
 def gatherSystemInfo(ctx):
+    import psutil
     system_info = {}
     system_info['bitness'] = platform.architecture()[0]
     system_info['machine'] = platform.machine()