[drcachesim] do not dump trace buffer on non-blocking syscalls

Today drcachesim's tracer dumps its per-thread buffers prior to each syscall just in case it blocks, to ensure we get timely output and timestamps for thread interleaving at key thread switch points. However, this can result in too many thread interleavings and very short buffer dumps. We could have a list of syscall numbers that never block and avoid dumping prior to those if we do not want to get into the more complex game of figuring out whether a syscall will block or not (esp without history after attaching).