Claude review: drm/v3d: Release indirect CSD GEM reference on CPU job free

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

Patch Review

**Analysis of the bug:** `v3d_get_cpu_indirect_csd_params()` calls `drm_gem_object_lookup()` at `v3d_submit.c:497`:

```c
info->indirect = drm_gem_object_lookup(file_priv, indirect_csd.indirect);
```

`drm_gem_object_lookup()` takes a reference on the GEM object, but nowhere in the teardown path was `drm_gem_object_put()` called to release it. This is a straightforward reference leak.

**Fix:**

```c
if (job->indirect_csd.indirect)
	drm_gem_object_put(job->indirect_csd.indirect);
```

The NULL check is appropriate: for non-indirect-CSD job types, `indirect_csd.indirect` will be zero-initialized (the job struct is allocated with `kcalloc` in `v3d_job_allocate`). It also covers the case where `v3d_job_init()` fails before the extension parsing reaches the indirect CSD path.

The placement in `v3d_cpu_job_free()` (the kref destructor from patch 1) is the right spot — it ensures the reference is dropped on both the scheduler path and the ioctl error path.

**No issues found.** Clean and correct.

---
Generated by Claude Code Patch Reviewer