Claude review: drm/imagination: Count paired job fence as dependency in prepare_job()

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

Patch Review

**Status**: Correct fix, Cc: stable appropriate.

This fixes the case where `job_count_remaining_native_deps()` failed to count the paired job fence as a native dependency because `to_pvr_queue_job_fence()` can't resolve it (the parent fence isn't assigned until submission). The fix manually counts it.

The `is_paired_job_fence()` helper is well-factored and immediately reused in the existing `pvr_queue_submit_job_to_cccb()` check:

```c
+static bool
+is_paired_job_fence(struct dma_fence *fence, struct pvr_job *job)
+{
+	/* This assumes "fence" is one of "job"'s drm_sched_job::dependencies */
+	return job->type == DRM_PVR_JOB_TYPE_FRAGMENT &&
+	       job->paired_job &&
+	       &job->paired_job->base.s_fence->scheduled == fence;
+}
```

The `WARN_ON(dma_fence_is_signaled(fence))` in the counting code is a reasonable sanity check - the paired job's scheduled fence shouldn't be signaled before submission.

One minor observation: the comment update `"backed by a UFO"` -> `"already backed by a UFO"` on `to_pvr_queue_job_fence()` is a good clarification to include here since it directly relates to understanding why the paired fence can't be resolved.

**No issues.**

---
Generated by Claude Code Patch Reviewer