Add core-sharded scheduling support to online drmemtrace analyzers

The new drmemtrace scheduler has been focused on offline use. It only interleaves separate input reader_t streams. Yet online mode uses a single reader_t for the entire process, so the scheduler does not do any scheduling of separate threads inside the analyzed process. This is a feature request to add that support, with the primary use case being simulators like dr$sim where we don't want either the as-traced schedule (-cpu_scheduling) nor the simplistic static scheduling inside dr$sim. Possibilities include ipc_reader_t multiplexing separate reader_t interfaces and not changing the scheduler; or changing the scheduler.