fix: use full-range field maps and field-aware kickoff positioning

[jaagut]

Field maps previously encoded grayscale values in 0–100. This migrates all nine localization maps to 0–255 and updates localization to normalize inverted byte intensities correctly. OccupancyGrid output remains 0–100, out-of-field score parameters retain their percentage semantics, and measurement marker colors use the normalized score directly.

Adds field.markings.center_circle.diameter to every field config. During opponent kickoff preparation (READY/SET), the central offense player uses circle radius + kickoff_striker_circle_margin as its distance behind the center. The margin defaults to 0.4 m. This applies only to normal/extra-time kickoff preparation with no other set play active. Other role positions and formation/set-play clearance retain their existing behavior.

Maps with saved generator inputs were regenerated as distance maps. Legacy maps without inputs were rescaled to preserve raster geometry; demo has no map. Every map retains its dimensions and exact zero-valued line mask. Existing goal-depth overrides remain intact. Circle diameters come from saved inputs or legacy raster geometry; demo declares no circle. The regeneration recipe is provided in this PR comment, rather than shipped as a tool.

Related upstream drafts:

• feat: make field map grayscale bounds configurable ros-sports/soccer_field_map_generator#4
• feat: export field geometry with center-circle diameter ros-sports/soccer_field_map_generator#5

Validation through Pixi:

• Built bitbots_localization, bitbots_parameter_blackboard, bitbots_blackboard, and bitbots_body_behavior for the map migration.
• Localization CTest passed for every byte intensity, occupancy-grid bounds/rounding, and unchanged out-of-map scoring.
• Revised behavior/blackboard package tests cover DSD parsing, mypy, and 17 kickoff regression cases: circle sizes, READY/SET, other roles, own kickoff, normal play, other set plays, and penalty shootout.
• All maps passed range/dimension/line-mask checks and reproduced byte-for-byte with the regeneration recipe.
• Changed Python files formatted and linted.

No robot or simulator runtime validation was performed.

[jaagut]

Regeneration recipe (kept here rather than shipped in the repository):

Run this from the repository root through pixi run -e default python /tmp/regenerate_field_maps.py, with the generator package built or its source on PYTHONPATH.

"""Regenerate distance maps from saved generator inputs.

Run through Pixi with the soccer_field_map_generator package on PYTHONPATH.
Legacy fields without generator inputs retain their existing map geometry.
"""

from pathlib import Path

import cv2
import numpy as np
import yaml
from soccer_field_map_generator.generator import generate_map_image

ROOT = Path("src/bitbots_misc/bitbots_parameter_blackboard/config/fields")


def main():
    for directory in sorted(ROOT.iterdir()):
        source = directory / "gui-config.yaml"
        if directory.name == "hsl_kid":
            source = directory / "hsl_kid.yaml"
        output = directory / "lines.png"
        if source.exists():
            parameters = yaml.safe_load(source.read_text())["parameters"]
            image = generate_map_image(parameters)
        elif output.exists():
            image = cv2.imread(str(output), cv2.IMREAD_UNCHANGED)
            if image.max() == 100:
                image = np.rint(image.astype(float) * 255 / 100).astype(np.uint8)
        else:
            continue
        if not cv2.imwrite(str(output), image):
            raise OSError(f"Could not write {output}")


if __name__ == "__main__":
    main()

Fields with saved generator inputs are regenerated; legacy maps are rescaled without changing geometry. Demo has no map. The full-range auxiliary binary images were already correctly encoded. Existing blackboard goal-depth overrides are preserved.