Fix C++ __guard implementation for ARM. The standard C++ ABI that most...

Fix C++ __guard implementation for ARM.  The standard C++ ABI that most platforms follow defines __guard to be 64 bits.  The existing implementation of libxx_cxa_guard.cxx follows this.  However, the 32-bit ARM C++ ABI defines it as 32 bits instead, and changes the meaning slightly so only the lowest bit is used. This matters because GCC creates guard symbols without regards to what libxx_cxa_guard.cxx says.  So on ARM, gcc allocates 4 bytes, but __cxa_guard_release writes 8 bytes, zeroing out another unlucky variable nearby. Fix it by special-casing 32-bit ARM in libxx_cxa_guard.