Don't blanket impl LenTextSize

[CAD97]

Instead, just provide copy-paste impls for somewhat commonly used std string containers.

[CAD97]

This does mean that smol_str either has to impl LenTextSize directly, or the user of both SmolStr and text-size has to use TextSize::of::<&str>(&SmolStr). IIRC, this is only one place in rowan, however, so this shouldn't be a big impact on ra.

[matklad]

The irony here is that if we made extension trait, and not TextSize::of, the method to create text sizes, then we'd get deref coercions for free. I do still feel that ::of design better. I even thing that of is better than of_char / of_str

[CAD97]

If only we could rely on the fact that char: Deref will never hold, then the following should be enough:

/// Text-like structures that have a text size.
pub trait TextLen {
    /// The size of this text-alike.
    fn text_len(&self) -> TextSize;
}

impl TextLen for str {
    #[inline]
    fn text_len(&self) -> TextSize {
        self.len().try_into().unwrap()
    }
}

impl TextLen for char {
    #[inline]
    fn text_len(&self) -> TextSize {
        (self.len_utf8() as u32).into()
    }
}

impl<D: Deref> TextLen for D
where
    D::Target: TextLen,
{
    fn text_len(&self) -> TextSize {
        self.deref().text_len()
    }
}

With this formulation the error isn't great but at least makes sense (rather than infinite recursion):

error[E0277]: the trait bound `(): std::ops::Deref` is not satisfied
  --> tests\constructors.rs:35:26
   |
35 |     let _ = TextSize::of(());
   |                          ^^ the trait `std::ops::Deref` is not implemented for `()`
   | 
  ::: D:\repos\rust-analyzer\text-size\src\size.rs:60:18
   |
60 |     pub fn of<T: TextLen>(text: T) -> TextSize {
   |                  ------- required by this bound in `text_size::size::TextSize::of`
   |
   = note: required because of the requirements on the impl of `text_size::traits::TextLen` for `()`

Though I do suppose that API doesn't help prevent TextSize::of(String).

[matklad]

bors r+