trace a subset of threads in drcachesim

This is a feature request to only trace some of the app's threads.

There are many headaches trying to have some threads running natively, so the plan is to run everything under DR and only disable tracing instrumentation.

For optimal performance we'd want to partition the code cache to have 2 copies of each fragment, one with instru and one without, with no explicit checks. That would take some work, so initially we'll probably go with a TLS check and skip the instrumentation for un-selected threads.