Claude review: media: meson: vdec: Fix memory leaks and lifetime of m2m device

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

Patch Review

**Good:** Moving `v4l2_m2m_init()`/`v4l2_m2m_release()` from per-session to per-device is the correct fix. The m2m device should be shared across sessions.

**Good:** Adding `v4l2_ctrl_handler_free()` in `vdec_close()` fixes a real leak shown in the kmemleak trace.

**Issue — error path in `vdec_open()` skips `v4l2_fh_del()`:**

The original `vdec_open()` calls `v4l2_fh_init()` and `v4l2_fh_add()` before returning success (lines 909-911 of the original). The new error path is:

```c
err_fh_del:
	v4l2_fh_exit(&sess->fh);
err_free_sess:
	kfree(sess);
```

The label says `err_fh_del` but only calls `v4l2_fh_exit()` — it never calls `v4l2_fh_del()`. Looking at the original code, `v4l2_fh_add()` is called **after** the point where `err_fh_del` would be jumped to (when `v4l2_m2m_ctx_init` fails), so `v4l2_fh_del()` may not be needed there. But the label name is misleading. However, if `vdec_init_ctrls()` fails, the code jumps to `err_free_sess` which skips the `v4l2_fh_exit()` entirely, and also leaks `sess->m2m_ctx` since `v4l2_m2m_ctx_release()` is never called. The error label when `vdec_init_ctrls()` fails should be `err_fh_del` (or a new label that releases the m2m ctx), not `err_free_sess`.

**Issue — `vdec_remove` ordering:** The `v4l2_m2m_release()` is called in `vdec_remove()` after `v4l2_device_unregister()`, which is correct, but `video_unregister_device()` should block until all open sessions are closed, so there shouldn't be an issue.

---
Generated by Claude Code Patch Reviewer