Make the eyes follow detected faces; start the face at power-up

hermes_face now subscribes to /vision/faces and glances toward the best face (a
recognised one first, then the biggest), smoothed, holding 2.5 s and drifting
back to centre when none is seen. Sleeping ignores gaze. Image x flips on the
display: the camera looks the way the robot faces, so a person on the image's
right is on the robot's right, the viewer's left.

Fix the pan sign carried over from the prototype: positive pan is the robot's
left (the MCP contract), which is the viewer's right on the display.

robot-face.service starts the face at boot and restarts it on failure, with a
private working directory because lgpio drops a notification pipe in its cwd and
systemd's default / is not writable. robot-cpu-performance.service holds the CPU
at full clock, which halves the face's CPU cost.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AFN8nRRCiDiMN7ChBpHJhs

ros/hermes_face/hermes_face/gaze.py

@@ -0,0 +1,31 @@
+"""Where the eyes should look, given where faces are in the camera image. No ROS imports."""
+from __future__ import annotations
+
+from typing import NamedTuple, Optional, Sequence
+
+
+class Target(NamedTuple):
+    cx: float          # face centre, pixels
+    cy: float
+    w: float           # face size, pixels
+    h: float
+    score: float = 1.0
+    name: str = ''     # who, if recognised
+
+
+def pick_target(faces: Sequence[Target]) -> Optional[Target]:
+    """A recognised face wins over an unknown one; otherwise the biggest, most confident face."""
+    if not faces:
+        return None
+    return max(faces, key=lambda f: (bool(f.name), f.w * f.h * f.score))
+
+
+def gaze_from_pixel(cx: float, cy: float, width: float, height: float) -> tuple[float, float]:
+    """Image position -> eye direction (+x the viewer's right, +y down), each in [-1, 1].
+
+    The camera looks the way the robot faces, so a person on the right of the image is on the
+    robot's right, which is the viewer's left on the display: x flips, y does not.
+    """
+    gx = (cx - width / 2.0) / (width / 2.0)
+    gy = (cy - height / 2.0) / (height / 2.0)
+    return -max(-1.0, min(1.0, gx)), max(-1.0, min(1.0, gy))