quickfix

Copying data dictionaries was very slow on Windows, which in turn made creation
of large numbers of sessions very slow.  Profiling showed that most of the time
was being spent in STL allocator code when calling the data dict copy
constructor.  The session factory now copies the data dict exactly once for
each session.  Interestingly, doing the copy on the heap instead of the stack
seems to have made the biggest perf improvement.  

Also, the Generator now attempts to use jruby instead of MRI ruby if it is
available.  This vastly speeds up code gen time (especially on Windows).