[PATCH] [lsan] Enable i386 support. Also rework the lit test configs.

[Alexander Potapenko]

  Can we be more strict in CanBeAHeapPointer on i386? E.g. say that an pointer must be aligned on 4, or point to the first X bytes of the allocated chunk?
  We may report some false positives this way, however.
  I think DrMemory people had an article about various heuristics for detecting heap pointers on 32-bit systems.

http://llvm-reviews.chandlerc.com/D1015