minor: Migrate (un)wrap_return_type assists to use SyntaxEditor

[Giga-Bowser]

Part of #15710 and #18285

This PR also includes some small fixes to SyntaxEditor, in particular a fix to the problem with affected_range outlined here.

I also updated unwrap_return_type to use placeholders when unwrapping None. I had tried to do this with the mutable tree API before and it was kind of a pain so it's nice to see improvements like this :)

[DropDemBits]

I can't add a comment to the file in question, but it would be good also add a test in syntax_editor::tests testing that token edits in a parent node won't be considered ancestors of edits in the child node:

#[test]
fn test_replace_token_in_parent() {
    let parent_fn = make::fn_(
        None,
        make::name("it"),
        None,
        None,
        make::param_list(None, []),
        make::block_expr([], Some(make::expr_unit())),
        Some(make::ret_type(make::ty_unit())),
        false,
        false,
        false,
        false,
    );

    let mut editor = SyntaxEditor::new(parent_fn.syntax().clone());

    if let Some(ret_ty) = parent_fn.ret_type() {
        editor.delete(ret_ty.syntax().clone());

        if let Some(SyntaxElement::Token(token)) = ret_ty.syntax().next_sibling_or_token() {
            if token.kind().is_trivia() {
                editor.delete(token);
            }
        }
    }

    if let Some(tail) = parent_fn.body().unwrap().tail_expr() {
        editor.delete(tail.syntax().clone());
    }

    let edit = editor.finish();

    let expect = expect![[r#"
        fn it() {

        }"#]];
    expect.assert_eq(&edit.new_root.to_string());
}

[DropDemBits]

Can also do

Suggested change

	|| (l.target_range().end() <= r.target_range().start())
	|| l.target_range().intersect(r.target_range()).is_none_or(|it| it.is_empty())

TextRange::intersect returns an empty range if the ranges are touching but not overlapping.

[Giga-Bowser]

I considered that but figured I'd rather just be straightforward with what's being checked especially since the ranges are sorted so there's no need to check if l starts inside of r

[DropDemBits]

Forgot to mention that the final tabstop snippet for unwrap_return_type should be after the last placeholder node instead of the last placeholder being considered the end of the snippet mode.