Claude review: driver core: platform: unify release path

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

Patch Review

**Correct, with careful refcount handling.** This moves `device_remove_software_node()` into the base `platform_device_release()` and removes the separate `platform_device_release_full()` introduced in patch 1.

The key subtlety is the `fwnode_handle_get()` for software nodes passed as primary fwnode:

```c
if (is_software_node(pdevinfo->fwnode))
    fwnode_handle_get(pdevinfo->fwnode);
```

This is necessary because `device_remove_software_node()` in the release path calls `kobject_put()` on whatever software node it finds. If the primary fwnode IS a software node (user-owned, not created by the platform code), the put would steal a reference the caller still holds. The extra get counterbalances this.

The placement before `platform_device_add_resources()` is correct: if any subsequent operation fails and we fall through to `err:`, `platform_device_put()` triggers the release callback, which calls `device_remove_software_node()`, which will consume the extra reference. Balanced.

The three cases work out:
- **Primary is software node, no secondary**: Extra get protects caller's reference; release puts it back.
- **Primary is non-swnode, secondary via pdevinfo->swnode**: No extra get needed; `device_remove_software_node()` removes the secondary swnode (added by this code).
- **Properties (managed secondary)**: Managed node is already cleaned by `software_node_notify_remove()` during `device_del()`; by the time `platform_device_release()` runs, `dev_to_swnode()` returns NULL, so `device_remove_software_node()` is a no-op.

**Verdict**: Correct.

---

---
Generated by Claude Code Patch Reviewer