[A] Add draw_detection func

[zephyr-sh]

✨ Summary

This PR introduces a new utility draw_detection() to the capybara/vision/visualization/draw.py module for rendering detection results on images, along with minor cleanups and shared constant reuse.

---

✅ Major Changes

1. New API: draw_detection()

Adds a high-level utility to draw labeled detection boxes (with optional confidence score) onto images.

Features:

• Box rendering:

  • Supports absolute and normalized box formats
  • Optional transparency via box_alpha

• Label rendering:

  • Auto-formatted confidence score (e.g., label 92.3%)
  • Text with semi-transparent background rectangle
  • Dynamic font sizing (~10% of box height, with min cap)

• Color selection:

  • Deterministic hashing for consistent per-class color
  • Based on golden ratio hue distribution

• Other options:

  • Optional font_path, text_size, text_color, text_bg_alpha
  • Auto thickness scaling based on image resolution

Example usage:

draw_detection(img, box=[x1, y1, x2, y2], label="cat", score=0.923)

---

2. Shared Font Path Constant

• Introduced DEFAULT_FONT_PATH as a shared constant
• Replaced duplicated string literals with reference to the constant

---

3. Guard for Font Download

• Updated logic to correctly check and download the font only if missing:

  if not (font_path := DIR / DEFAULT_FONT_PATH).exists():

---

🧠 Motivation

• Previous APIs such as draw_box() and draw_text() were too low-level for rendering detection results in a single call.
• This utility abstracts away box coordinate handling, color consistency, text layout, font sizing, and blending.
• Provides an out-of-the-box tool to visualize detection results for debugging, demos, or evaluation.

---

🖼️ Visual Behavior

• Consistent box color per label: ensures same class gets same hue
• Readable text: white font with semi-transparent background
• No overlap or cutoff: clamps background rectangle to image bounds
• Sensible defaults: line thickness and font size scale with resolution

---

🧪 Test Coverage

This utility is self-contained and built upon existing well-tested primitives:

• prepare_img, prepare_box
• cv2.rectangle, ImageFont, ImageDraw

Optional follow-up PR can add unit tests using synthetic inputs to assert image output properties.

---

🔄 API Summary

def draw_detection(
    img: np.ndarray,
    box: _Box,
    label: str,
    score: float | None = None,
    color: tuple[int, int, int] | None = None,
    thickness: int | None = None,
    text_color: tuple[int, int, int] = (255, 255, 255),
    font_path: Union[str, Path] | None = None,
    text_size: int | None = None,
    box_alpha: float = 1.0,
    text_bg_alpha: float = 0.6,
) -> np.ndarray

---

✅ Checklist

• draw_detection() implemented and documented
• Color hashing via label for class-consistent visuals
• Shared font path constant extracted
• Compatible with normalized boxes
• Layout clamped to avoid text overflow