feat: add clear api

[omfj]

Adds a SoAVec::clear() API.

This deallocated all the items inside of the SoAVec and sets the length to 0.

Moved the contents of the Drop-impl to a function called deallocate_items so that the same logic is used for clearing/dropping the items and dropping the SoAVec.

Closes #20.

[aapoalas]

issue: There's a subtle memory-safety problem hiding in here.

What happens if some drop of T or call to T::TupleRerp::drop_in_place panics inside the clear call?

1. Let's assume we've already dropped some T's in the SoAVec (otherwise the problem will not show itself).
2. A panic occurs inside of deallocate_items, meaning that we leave this function without performing set_len(0).
3. Regardless of if the panic is caught above or goes on to crash the program, the Drop impl of our SoAVec will get called: it starts dropping the items anew from the beginning.
4. Since we already dropped some items in the SoAVec in step 1, step 3 now performs double-free and we are in trouble.

Take a look at how Vec::clear is implemented; we'll need to follow its internal logic to make this safe.

[omfj]

I tried to look at how Vec::clear worked, but I didn't really understand what would be the equivalent of ptr::drop_in_place(elems).

[aapoalas]

Right so; right now we do not have a full equivalent to ptr::drop_in_place. We do have the T::TupleRepr::drop_in_place which works if !T::MUST_DROP_AS_SELF, but for structs that require dropping as Self we simply need to read the data out of the SoAVec, turn it into a T using T::from_tuple, and drop that.

So, effectively it's the full equivalent is deallocate_items except that deallocate_items currently uses self.len(). If we simply take len as a parameter (or perhaps all the ptr, cap, and len values) then we can reuse the function both here and in Drop; in Drop we don't need to call set_len as we're already dropping the SoAVec: if a panic happens inside Drop then it will not be retried.

In clear we'll have to read len, then set_len(0), and then call deallocate_items with the stored len value. That way even if deallocate_items panics, dropping the SoAVec will see a 0 length and will not try to re-drop the items.

[omfj]

Ok, hopefully I understood. I renamed the function to better match the one used in the actual Vec::clear too.

[aapoalas]

LGTM!