`TimeMgr::Clock` has no way to detach an `Alarm`, risking dangling pointers

[andrewdnolan]

Clock::attachAlarm stores a raw Alarm* in an internal vector, but there is no detachAlarm/removeAlarm counterpart anywhere in TimeMgr. Several classes (IOStream, TimeStepper, SfcCoupling) attach Alarm members to a shared Clock and later destroy themselves via erase()/clear() without detaching their alarms first. If the Clock outlives the destroyed object and advance()/setCurrentTime() is called, it dereferences a dangling Alarm* (use-after-free).

This issue came up when I had copilot review #458 locally. It's a bit out of my wheelhouse C++ wise, so very open to push back here.

[philipwjones]

It's unlikely to occur in practice - can't think of a use case for detaching an alarm other then on shutdown, but I can add it and a proper cleanup for completeness and good practice.