support clean calls with stack padding and memory opnds

_From [bruen...@google.com](https://code.google.com/u/109494838902877177630/) on February 19, 2014 15:45:39_

We have not hit this before because we pass -mpreferred-stack-boundary=2 on
Linux.

On 32-bit MacOS, though, we have stack alignment and we have args passed on
the stack. We need to update insert_parameter_preparation() to handle this
situation and have it procure some scratch registers in a safe manner.

_Original issue: http://code.google.com/p/dynamorio/issues/detail?id=1370_