Add image data representation for CNN backbones

[sevmag]

Summary

Second in the series of PRs being carved out of #813 (see the split tracking comment for the plan). This is the foundation for the CNN work — it adds an image-shaped data representation analogous to GraphDefinition, with no model code yet. PRs 3–4 (the actual CNN models) will build on this.

What's new

• ImageRepresentation base class — analogue of GraphDefinition for image-shaped inputs. Builds per-channel image tensors from raw pulse data using a GridDefinition that maps each DOM to a voxel.
• GridDefinition — abstract mapping from DOM string/dom_number to a 3D grid index. Two concrete subclasses:
  • IC86GridDefinition — IceCube IC86 array (main array + upper/lower DeepCore).
  • ExamplePrometheusGridDefinition — Prometheus example geometry.
    Mapping tables live in cnn_mapping_tables.py as Python literals (no parquet files).
• IC86Image and ExamplePrometheusImage — concrete image representations bundling a detector and its grid definition.
• New TEST_IMAGE_DIR / TEST_IC86*_IMAGE constants and .npy test fixtures for IC86 main array + DeepCore sub-arrays.
• Unit tests for GridDefinition and ImageRepresentation (3 tests, all passing).

What's not in this PR

• No CNN model code — that's PR 3 (IceCubeDNN) and PR 4 (LCSC).
• No training example — that's PR 5.

Notes for review

• All new files; only two existing files are touched (src/graphnet/constants.py to add 4 test-data path constants, and data_representation/__init__.py to re-export the new image classes).
• The previous parquet-based mapping table approach was dropped in favour of inline Python literals (see cnn_mapping_tables.py).

Test plan

• Pre-commit clean (black, flake8, docformatter, pydocstyle, mypy, EOL/whitespace)
• pytest tests/models/test_grid_definition.py tests/models/test_image_representation.py — 3 passed
• CI green

Split from #813.

🤖 Generated with Claude Code