Bug to fix: drmgr counters being nonzero when exit, leading to corrupted drmemtrace traces in subsequent execution

Created by: Louis-Ye

It looks like drwrap doesn't clean up all the resources when calling drwrap_exit() in client. In might cause potential problems in clients when DR is actively invoked and stopped by the same user application several times in one run. Here is how I reproduce the problem in drcachesim client test case:

1. Call drwrap_init() inside clients/drcachesim/tracer/tracer.cpp:drmemtrace_client_main
2. Call drwrap_exit() inside clients/drcachesim/tracer/tracer.pp:event_exit
3. include "drwrap.h" in clients/drcachesim/tracer/tracer.pp, add use_DynamoRIO_extension(${name} drwrap${ext_sfx}) under clients/drcachesim/CMakeLists.txt:configure_DynamoRIO_client
4. Build with test cases
5. ctest -R .*burst_static will fail 2 out of 3 times (test case: tool.drcacheoff.burst_static).