Fix analyzer clip framerate

## Summary

`render_timeline_clip()` in `artemis/utils/video.py` renders every analyzer clip at ~25fps instead of the requested `fps` (default 15), silently duplicating frames. Since the whole point of this function is to keep a frame index mappable back to a real recording timestamp, the mismatch breaks that guarantee on every analyzer clip the agent renders.

## Root cause

Each input segment's ffmpeg filter chain sets an explicit `fps={fps}`, but the final encode command never sets an explicit output frame rate (`-r`). Without one, ffmpeg has no framerate to encode the muxed output with and falls back to a default of 25fps, inserting duplicate frames (via PTS) to preserve the real wall-clock duration:

```
frame= 77 fps=0.0 q=28.0 Lsize= 4KiB time=00:00:03.00 bitrate=10.6kbits/s dup=31 drop=0
```

A 3.0s window that should produce ~45 frames at 15fps instead produces 77 (~40% duplicate frames). Duration is correct; encoded framerate is not.

## Impact

`UnifiedMobileController.render_timeline_clip` (`artemis/controllers/unified_controller.py:408`) is the only caller and runs on the production path — every analyzer clip the agent renders during a task goes through this. Any downstream logic assuming `frame_index / fps ≈ recording_time` (which is exactly what this function's docstring promises: *"so `start_time + frame_offset` still names the true recording time"*) is working against a clip whose actual framerate silently diverges from `fps`.

## Reproduction

This is caught by an existing, currently-failing test on `main`:

```
tests/unit/test_unified_controller_video.py::test_analyzer_clip_keeps_timeline_time_across_restart_gap
FAILED - assert 43 <= len(frames) <= 47
AssertionError: assert 77 <= 47
```

Reproduces the same way outside pytest by calling `render_timeline_clip()` directly and reading frame count with `cv2.VideoCapture`.

## Fix

Pass `-r {fps}` to the output stage of the ffmpeg command so the encoder gets an explicit framerate instead of guessing:

```diff
"-filter_complex", ";".join(filter_parts),
"-map", "[outv]",
+ "-r", str(fps),
"-c:v", "libx264",
```

Verified: frame count drops from 77 → 46 (within the test's expected 43–47 range); the previously-failing test now passes, and the full `tests/unit/test_unified_controller_video.py` file (22 tests) passes. `ruff format --check`, `ruff check`, and `pyright` are clean on the changed file.

Start in artemis/utils/video.py at render_timeline_clip() and inspect the final ffmpeg output stage, then read the caller at artemis/controllers/unified_controller.py:408. Run tests/unit/test_unified_controller_video.py::test_analyzer_clip_keeps_timeline_time_across_restart_gap and the full test file; done means the expected frame count passes and the stated format, lint, and type checks remain clean.

Written by the indexing model from the issue text.