provide faster dr_try_setup() that doesn't allocate memory

_From [bruen...@google.com](https://code.google.com/u/109494838902877177630/) on December 07, 2011 22:02:20_

current version allocates memory to provide an abstraction barrier and
allow for freedom of impl changes w/o breaking compatibility.
the memory allocation is showing up in profiling of Dr. Memory.
we should provide both and document that the fast one may not remain
compatible, in the spirit of fast IR (issue #622)

_Original issue: http://code.google.com/p/dynamorio/issues/detail?id=635_