Introduce a Clearable trait to support unsized types

[burdges]

Introduce a Clearable trait to support unsized types. Addresses the most concerns in #3 as well, including the warning by @bluss that assignment might not zero everything #3 (comment)

I decided to implement Clearable for all Sized+Copy types because Shared<T> is the only dangerous Copy type I could find. there are many pointer types that satisfy Default but Shared<T> does not, so we can avoid all current pointer types if we used impl Clearable T where T: Copy+Default instead.

I added this in ccb563e but reverted it in 88b4c4f because Shared<T> does not seem so dangerous right now. I'd hope Rust gets type-level numerics before any new dangerous Shared<T> abstractions and any such abstractions are likely to satisfy Default anyways.

We do drop a zeroed object as a result of reverting the Default bounds, but afaik the worst case would be dropping a zeroed Shared<T>, whish does nothing itself. If anything worse happens, then you can revert the revert 88b4c4f of course. :)

[burdges]

We should probably add some tests for unsized types besides the [T] in a Vec<T>, like maybe the [T] in a Box<[T]>.

[burdges]

Added impls for AsRef and AsMut. I'll send a separate pull request with my thoughts on Borrow and BorrowMut.

[burdges]

I should likely make this AsRef and Borrow stuff a separate pull request, as it's maybe a distraction here.

Just fyi, I tried two tricks to make ClearOnDrop behave kinda like it took a Borrow, but both failed : https://github.com/burdges/clear_on_drop/tree/borrow_fail/src https://github.com/burdges/clear_on_drop/tree/owned_fail/src

[burdges]

Just changed how unsized arrays are handled to hopefully deal with #6 (comment) I have not tried to verify that hide_mem really does not work on unsized types. Also I have not yet removed the ?Sized bounds this pull adds in hide_mem.rs, which probably should happen.

I maybe delt with #7 (comment) too, which required adding Default back.

[burdges]

I'm not confident what should be marked unsafe in the Clearable trait. In particular, there is less need for clear() to be unsafe now that it no longer leaves T in an unusable state. I think marking the Clearable trait itself as unsafe now more important since you cannot implement it yourself without using something like hide_mem.

Also ClearInPlace would be another possible name for Clearable. There is no need for a flashy name if almost nobody will be implement it themselves or use it as a constraint. :)

[cesarb]

I went in a slightly different direction (inspired by this pull request) with commit 517686e, and now allow [T] with the possibility of the user adding his own DSTs by implementing the InitializableFromZeroed trait.

With that change, this pull request will become hopelessly tangled, so I'll close it now; please open a new one based on the current master if there are still things to change. I'll probably release 0.2 in a few days.