Otherwise, you need a way for the cleanup code to dispose of the resources and leave the system in a consistent state before crashing.
Ok, its obviously good practice to develop your software in a way that a random outage doesn't leave anything in an inconsistent state, but a lot of software fails to do this. At the very least, you may end up with something like when a server crashes and cannot restart because the port is still considered in use by the OS.
Imagine the customer error report or imagine its a critical system and it starts making mistakes...
For most software smoke testing heavily is enough to make it super rock solid. I know that most software does not do this - just using a web browser or a smartphone makes it painfully obvious that even the big software houses have some seriously shoddy practices and that testing gets seriously neglected (it may be that its impractically big... i stuggle to buy that tbh)
I really strongly disagree with this. Better for it to crash if you expect this memory to always have been allocated. This way you can fix your bug instead of putting your app into some potentially unexpected state... if allocations fail it doesn't make sense to just 'carry on anyway' in many situations.