Add drcachesim global trace duration limit that does not kill the process

To limit the trace size for cases where adding start/stop calls into the code is not feasible or convenient, we have the -exit_after_tracing and -max_trace_refs options. However, the former kills the process, causing various job control and post-trace-action problems, while the latter is a per-thread limit, which is not useful for limiting the overall window.

We want either a -detach_after_tracing, which requires a client-triggered detach which is #2644, or a -max_global_trace_refs.