Claude review: rust: pci: add physfn(), to return PF device for VF device

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

Patch Review

This is the most interesting patch. The safety argument is well-constructed:

```rust
+    pub fn physfn(&self) -> Result<&Device<device::Bound>> {
```

The method is on `Device<device::Bound>` (not `Device<Core>`), which means it's callable from any bound context. Three defensive checks ensure soundness:

1. `is_virtfn()` — confirms the union field `physfn` is valid
2. `pf_drv.is_null()` — PF has a driver bound
3. `!(*pf_drv).managed_sriov` — the binding guarantee is upheld

The runtime check for `managed_sriov` on the PF's driver is essential for the case where SR-IOV was enabled by a C driver that hasn't opted in. This correctly returns `EINVAL` rather than producing an unsound reference.

**One consideration**: All three error cases return `EINVAL`. For debugging, it might be helpful if `!is_virtfn()` returned a different error (e.g., `ENODEV`), but this is minor.

The reference lifetime is tied to `&self` (the VF device), which is sound: the VF can only exist while the PF is bound (guaranteed by `managed_sriov`), so the PF reference is valid for at least as long as the VF reference.

The commit message correctly notes the dependency on "rust: driver: drop device private data post unbind" for handling the pathological case of a PF driver re-enabling SR-IOV in `unbind()`.

---
Generated by Claude Code Patch Reviewer