Threads may access destructed global `DbgCtl` objects on shutdown

[JosiahWI]

The issue was originally reported by TSan as a shutdown race. That report is easily reproducible, but I don't have a TSan build right now to do so. There were concerns it could be a false positive, but ASan happened to actually catch it in the wild on CI. This does not mean definitively that it isn't a false positive, but as TSan rarely has false positives (except in certain cases of which I believe this is not one) and both tools detected it, I think it's reasonable to presume this is a real race unless it can be proven otherwise. Here is the ASan report:

==6586==ERROR: AddressSanitizer: heap-use-after-free on address 0x604000001178 at pc 0x0000004ab418 bp 0x7ff4f57e1cc0 sp 0x7ff4f57e1cb0
READ of size 1 at 0x604000001178 thread T1 ([ET_NET 0])
    #0 0x4ab417 in DbgCtl::on() const ../include/tsutil/DbgCtl.h:111
    #1 0x725570 in PollCont::do_poll(long) ../src/iocore/net/PollCont.cc:89
    #2 0x71d2bc in NetHandler::waitForActivity(long) ../src/iocore/net/NetHandler.cc:362
    #3 0xd4b8c2 in EThread::execute_regular() ../src/iocore/eventsystem/UnixEThread.cc:307
    #4 0xd4bddd in EThread::execute() ../src/iocore/eventsystem/UnixEThread.cc:358
    #5 0xd48a44 in spawn_thread_internal ../src/iocore/eventsystem/Thread.cc:75
    #6 0x7ff4f9db21c9 in start_thread (/lib64/libpthread.so.0+0x81c9)
    #7 0x7ff4f9a0d8d2 in __GI___clone (/lib64/libc.so.6+0x398d2)

This is likely very nontrivial to fix because of the difficulty of managing shutdown of global objects. Joining all threads before shutdown may be one possible fix.