Add RAII resource wrappers

[psychon]

This PR implements #78 for the core protocol: This adds e.g. x11rb::wrapper::WindowWrapper which, well, wraps a window ID (u32). In its Drop implementation, this calls destroy_window, thus preventing "window leaks".

I am not sure how much I really like this, but here it is in the form of the PR.

Also, there is way too much hand-written code in this PR. I guess I'd prefer if the code generator generated the necessary code for these wrappers. But today is not the day for that...

[eduardosm]

I'm not sure how this will scale for large projects, where XIDs might be stored in structs. Those structs would need have a connection lifetime parameter. Additionally, the same struct wouldn't be able to own both the connection and a wrapped XID, because it would lead to self-references.

Additionally, I'm not sure I'm happy with sending data over a socket inside a drop, because drop can be called during a panic unwind (in that situation, you probably don't care about protocol state a want to bail out as fast as possible).

Anyways, as long as wrappers are optional, I'm fine with it.

[psychon]

Yeah, this is definitely not useful for everyone.

Why I did this: I recently implemented paste (ICCCM selection transfers) "just because". For the transfer, the code creates a window that is destroyed at the end. I ended up writing another version of WindowWrapper for this to make sure this temporary window is always destroyed: linebender/druid@48e5a21#diff-d692afae7592991d6306b5f4fde771c27317f806d752bcf713370b27356ce1f7R157

There is also code like this: https://github.com/linebender/druid/blob/f74a7120fa724b2d6dba88573942aa9864c16b94/druid-shell/src/platform/x11/window.rs#L290

The line in question is a create_window request. After this request, there are 10 ?. Most of them would lead to a "window leak" if they return. I feel like code like this would really benefit from "something RAII".

because drop can be called during a panic

Hm, yeah... So, what do you think? I described how I ended up writing this PR. Am I approaching this completely wrong or could this still be useful?

I am not sure how much that drop-during-panic is a problem. If it is, the code could check whether a panic is in progress and not free the window in this case, but this seems like a weird special case to me...

Anyways, as long as wrappers are optional, I'm fine with it.

Yeah, this definitely cannot become the only possible way. Self-referential structs and all. I guess I'd want to add something like this for "everything" (not only core-protocol). Writing this by hand is way too cumbersome and proc_macro also seems complicated, so I guess I'd want to eventually move this into generator/. Something like a hardcoded list inside the code generator or perhaps via extra XML: #589. But that shouldn't change the public API of these wrappers, so this PR is mostly about that (and whether adding this is a good idea at all).

---

Note to self: Before merging this, there should also be some into_window() method that consumes the WindowWrapper and returns the raw underlying window. That would be helpful for the code in druid, I think. I'll add that now.