Implement UnwindSafe by never returning on panic

[fabianfreyer]

Don't return objects to the pool on panic, since we can't be sure whether they were the source of the panic.

See-also: #139

[bikeshedder]

I think you accidentally mixed the managed and unmanaged implementations.

The Drop implementation of deadpool::managed::Object needs to be changed and the UnwindSafe marker needs to be added to the deadpool::unmanaged::Pool for both implementations to be deemed unwind safe.

[fabianfreyer]

Hmm, I guess both should have the marker then?

[bikeshedder]

I'm just revisiting this as part of a overall repository clean-up. Not returning objects on panic is a good idea for the managed pool. For the unmanaged pool hower this could cause some rather pesky side effects. It's unmanaged for a good reason: You create all the objects you need and then just use them forever. If a panic doesn't crash the application altogether it might end up with an empty pool waiting forever.

One way I could think about was the following:

• managed pools don't return the object if the code panicked. Though they still need to make sure to notify the pool about the slot being available!
• unmanaged pools should probably enter a panicked state and all future calls to get should panic as well. Maybe make this configurable somehow?

It's just my gut feeling that discarding objects on panic only ever makes sense if panics are handled and then you really need to make sure that the pool gets refilled. I can imagine only very few users would expect that to be the case.