treat client meta-code differently than app code wrt system calls and exceptions

_From [derek.br...@gmail.com](https://code.google.com/u/117968039472581148324/) on February 24, 2009 14:47:14_

this was PR 213600

We may want to split client interface code into two modes:
acting-as-application and acting-at-meta-level. If the client is
transforming the application code, then system calls or exceptions in
client-introduced code should be treated as any other app code. However,
for client code that is merely observing and analyzing, we should provide
support for treating that like we treat hotpatch code or DR code: don't
interrupt control for syscalls (if meta-client-code is acting while a
thread is UNDER_DYN_HACK or for some reason native_exec) and don't throw
exceptions back to the app (xref hotpatch exceptions, try-except support
for clients w/o them mucking w/ application exception handlers, etc.)

A good example here is for clean call args that reference memory, where the
client may prefer to receive the fault itself rather than it being treated
as an app exception (xref PR 302951).

_Original issue: http://code.google.com/p/dynamorio/issues/detail?id=55_