Add functions for writing zeroed bytes to &mut impl Zeroable and &mut [impl Zeroable]

[thomcc]

Two reasons this functionality is worth having:

1. Some cases in FFI want a struct required input to be cleared with memset, because they expect to add fields to the end of a struct and use sizeof(TheStruct) to indicate which fields exist on the type.

   That is: a new field could be added to the end of a struct without changing the size if the bytes of that field were previously tail padding -- in this case if all bytes were not zeroed in this manner, the uninit bytes in the field may be interpreted as having meaning.

2. *blah = T::zeroed() will end up having a few redundant copies of T on the stack in debug builds which could cause various problems.

I didn't add these as methods on Zeroable intentionally, but don't care much about it. If you'd prefer them there or something else I'm fine with it.

I'm pretty sure I didn't miss anything in terms of correctness here, although I had initially forgotten that Zeroable didn't imply Copy, so a bit of scrutiny wouldn't be out of place to ensure I'm not adding a problematic API.

[thomcc]

I decided supporting non-copy types was actually useful because sometimes code doesn't bother with a #[derive(Copy, Clone)]. Also Cell and stuff.

But a lot of them won't need dropping (so this should still be fast) and it isn't really that hard to handle the ones that do.

[thomcc]

Also while this does clobber the provenance of any pointer it overwrites with zero bytes, that is true for writing ptr::null() too (as null() has no provenance).

(This is probably not news tho, I had just not realized it. Presumably this is part of why it's fine for raw pointers to have a Zeroable impl)

[a1phyr]

This will lead to a double-drop if drop_in_place panics. You should probably call write_byte in a drop guard instead.

[Lokathor]

This is true. Alternately, do we want to require Copy after all?

[a1phyr]

I don't see any reason to limit to Copy types if there is a zero-cost and correct way to support all types

[Lokathor]

Alright, just checking i hadn't forgotten anything i guess

[thomcc]

hmmm, lets just make these require Copy

[thomcc]

Oh, sure, we can do it in the drop guard.