Make `Query<&T>` return `Ref<T>`

[alice-i-cecile]

          A potentially more controversial option is just to have `&T` return `Ref<T>` instead. Though that might require some benchmarking to ensure there's no perf regression in doing so.

Originally posted by @james7132 in #7306 (comment)

A nice thing about making &T return Ref is that it would make the ReadOnly assoc type on &mut T have an "obvious answer", with this PR it could realistically be &T or Ref. I think the only downside that Mut has that Ref would also have is the fact that you cant pattern match on it, the other issues (shoving mut everywhere and borrowck annoyances from disjoint borrows) dont apply to Ref which is cool.

Boxy's followup.

This would align us further with the behavior of &mut, and ensure that change detection behavior is always available to users.

[Guvante]

There is also #6659 which is going in the opposite direction.

However on second thought both of these could be used together. Defaulting to change detection which returns Ref<T>/Mut<T> and if you disable change detection you get &T/&mut T could be pretty straightforward.

Also I don't know if it is good or bad but with simpler types it would be more common to disable change detection when it wasn't used.

[Guvante]

It sounded interesting so I whipped together a first pass example of what I said. I can spend some time making the PR more thorough later if this path looks like an productive one.

[alice-i-cecile]

Yeah, I rather like this set of changes: I'd be happy to see this proceed.

[cart]

Left some thoughts here: #7344 (comment)