1.0.4.18: trap handling cleanup continues

 * handle_trap is even better then maybe_handle_trap: if we can't
   handle the trap sanely pass the buck to unhandled_trap_error
   instead of generic signal handling routines.

 * implement Windows trap handling on top of handle_trap, not
   sigtrap_handler: some copy-paste from x86-arch.c later Windows is
   now almost free of the "signal handling" remnants.

 * Replace the bogus lossage messages saying
   "fake_foreign_function_call fell through" with things that actually
   make sense.