Add overexposed_frame_pct to frame stats from the luma values already computed #33

Current behavior

The frame instrument already computes per-frame average luma for every frame (src/hflow/ffmpeg/_instrument.py:151), but only three aggregates are reported (_instrument.py:181-183): luma_avg_mean, luma_avg_min, and luma_avg_max.

Blackout has a dedicated metric (black_frame_pct, _instrument.py:41 and :178), but the symmetric failure (blown highlights from bad exposure, IR flood, or sun in frame) has no counterpart. luma_avg_max alone cannot distinguish one bright frame from a thousand.

Proposed change

• Add overexposed_frame_pct to the FrameStats dataclass (_instrument.py:34-46), computed as the percentage of frames whose average luma is at or above a threshold.
• Expose the threshold as a parameter on frame_stats (for example, bright_luma_threshold: float = 235.0), matching the module's rule that thresholds are user-owned.
• Compute it from the existing luma_avg_values list: no new ffmpeg filter and no second decode pass.
• Surface the new value as a measurement in camera_frame_stats (src/hflow/checks.py:189), next to black_pct.

Definition of done

• frame_stats reports overexposed_frame_pct in the 0.0 to 100.0 range with the same denominator as black_frame_pct.
• camera_frame_stats records the measurement per camera topic.
• Extend tests/test_ffmpeg.py with a bright-video case (the existing fixtures show how the black-segment case is synthesized).

Validation

uv run pytest -q
uv run ruff check --fix && uv run ruff format
uv run ty check