update native_exec to properly swap the signal mask, TLS, etc.

The native_exec code is not using os_thread_{,not_}under_dynamo() which means it is not using the fixes and changes from the past year or so: swapping the signal mask (#2337 (closed)), clearing the TLS (#2089 (closed)), etc.

It looks like adding the os_thread_{,not_}under_dynamo() calls and the os_thread_re_take_over() to get the TLS and dcontext back are straightforward for default options, so I will go ahead and do it, but complex for -native_exec_opt as they need to be in generated code.