Contributor architecture

Start in src/chimera_ml when tracing a change:

• core defines containers, configuration loading, and registries;
• training/builders translates named YAML sections into runtime objects;
• training/trainer owns optimization, evaluation, callbacks, and logging;
• data normalizes dataloaders and creates masks;
• inference builds and executes step graphs;
• plugins.py imports built-ins and delegates entry-point discovery.

Tests are organized by subsystem under tests/: core, data, models, losses, metrics, callbacks, logging, inference, plugins, CLI, and training. Examples are separately packaged projects and are part of the integration surface.

When adding a user-facing component, update its registry registration, tests, example/config if appropriate, API reference, and both language trees.