Claude review: rust: types: introduce ForLt base trait for CovariantForLt

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

The trait split is well-designed:

```rust
pub trait ForLt {
    type Of<'a>: 'a;
    unsafe fn cast_ref_unchecked<'r, 'short: 'r, 'long: 'short>(...) -> ...;
}

pub unsafe trait CovariantForLt: ForLt {
    fn cast_ref<'r, 'short: 'r, 'long: 'short>(...) -> ...;
}
```

`ForLt` is not `unsafe` to implement (correct — it makes no variance guarantee), while `cast_ref_unchecked` is `unsafe` to call. `CovariantForLt` is `unsafe` to implement (the implementer guarantees covariance) with a safe `cast_ref`.

The `ForLtImpl` helper struct correctly omits the `const N: usize` parameter since no covariance proof needs to be embedded:

```rust
pub struct ForLtImpl<T: ?Sized, WF>(PhantomData<(WF, T)>);
```

The blanket `ForLt` impl on `CovariantForLtImpl` correctly covers all `N`:

```rust
impl<T: ?Sized + for<'a> WithLt<'a>, WF, const N: usize> ForLt for CovariantForLtImpl<T, WF, N> {
    type Of<'a> = <T as WithLt<'a>>::Of;
}
```

This is needed so that `CovariantForLt: ForLt` is satisfied.

The macro refactoring (`resolve_hrt`, `ty_static` shared helpers) is clean. The new `for_lt()` macro function correctly skips the covariance `Prover` pass.

**Minor observation:** `CovariantForLt::cast_ref` still contains its own `transmute` rather than delegating to `unsafe { Self::cast_ref_unchecked(long) }`. This is fine for performance (both are `#[inline(always)]`), but using the unchecked version would make the relationship between the two methods more explicit. Not worth a respin.

---
Generated by Claude Code Patch Reviewer