Claude review: rust: drm: add base GPUVM immediate mode abstraction

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

Patch Review

The foundational patch introducing `GpuVm<T>`, `DriverGpuVm` trait, and `UniqueRefGpuVm<T>`.

**Positive:**
- The `AlwaysRefCounted` impl correctly delegates to `drm_gpuvm_get`/`drm_gpuvm_put`.
- `vm_free` callback properly uses `container_of!` + `KBox::from_raw` for cleanup.
- The `UniqueRefGpuVm` invariant (at most one per `GpuVm`) is well-documented.

**Nit — error bound on `new()`:**
```rust
pub fn new<E>(
    ...
) -> Result<UniqueRefGpuVm<T>, E>
where
    E: From<AllocError>,
    E: From<core::convert::Infallible>,
```
The `E: From<core::convert::Infallible>` bound seems unnecessary — every type trivially converts from `Infallible`. This is likely an artifact of `try_pin_init!` macro expansion requirements, but it reads oddly.

**Send/Sync safety comments are thin:**
```rust
// SAFETY: The GPUVM api does not assume that it is tied to a specific thread.
unsafe impl<T: DriverGpuVm> Send for GpuVm<T> {}
// SAFETY: The GPUVM api is designed to allow &self methods to be called in parallel.
unsafe impl<T: DriverGpuVm> Sync for GpuVm<T> {}
```
These should note that `T` itself must be `Send`/`Sync` or explain why it doesn't matter. Currently the `DriverGpuVm: Sized` bound doesn't require `Send + Sync`, but `GpuVm<T>` is unconditionally `Send + Sync` regardless of `T`. The `data` field is behind `UnsafeCell` and only accessible through `UniqueRefGpuVm`, so `Sync` for `GpuVm` is reasonable, but the `Send` impl should require `T: Send` — otherwise a non-Send `T` could be moved across threads via `ARef<GpuVm<T>>`.

---
Generated by Claude Code Patch Reviewer