Enhance benchmark and Cortex modules with new training utilities and improved state management. Update README with example output for Llama-3.2-1B and add training CLI for Cortex module tuning. Refactor scoring functions to reset Cortex state between examples and ensure consistent output. Modify task handling to ensure proper formatting of input data.
Refactor device handling in benchmark and cortex modules to use resolve_torch_device function for improved device selection. Update README with benchmark table formatting and additional device options.
