Allow `FnMut` in blocks?

[madsmtm]

Upstream issue: SSheldon/rust-block#12.

Should we allow Block<dyn FnMut()>, and how would we do it safely?

At the very least, we'd have to prevent any form of Block::copy/RcBlock::clone on these, as that could lead to multiple mutable references being present. In particular, it is difficult to prevent re-entrancy if the block is shared.

This decision is blocked on a resolution in #563, though we might still decide to allow FnMut in block2, even if we don't allow it in objc2, as most (non-@escaping, non-@Sendable, see #572) blocks (probably) really are safely mutable.

See also #168 (comment) for discussion on why FnOnce is impossible, we should document that somewhere more visible when tackling this issue. Or maybe add RcBlock::once, similar to wasm-bindgen?

[madsmtm]

I've documented how to use FnOnce and FnMut in 19ec1c4, still a little bit unsure whether to unsafely expose FnMut somehow, though I don't currently think it'd be a good idea.

[madsmtm]

Yeah, I'm not going to expose FnMut blocks, since how to do so depends on a lot of factors, and it's difficult to write a correct wrapper.

The user should use RefCell or Cell themselves, as this both makes the drawbacks more clear, and allows them to choose the model that works for their use-case. And if you really want unsafe FnMut blocks, they can be implemented on top of Fn blocks with UnsafeCell.