[cmccomb]

Fresh installations of 0.1.2 can select incompatible Bincode and SmartCore versions, and malformed script data can panic the host. This prepares 0.1.3 with compatible dependency ranges and validated training/prediction inputs. It also prevents the reproduced underdetermined-linear-regression panic and reports conversion, serialization, backend, and prediction-overflow errors through Rhai.

The public train/predict signatures and algorithm defaults are preserved. Integer and mixed numeric regression inputs are supported. The README is reduced to 50 lines, with detailed input rules and development checks in linked documents.

Validation:

• 17 behavioral tests, two generated Rhai examples, and three Rust doctests pass with metadata; default-feature tests also pass.
• The suite passes against minimum-supported Rhai 1.16.2.
• Formatting, strict Clippy, documentation, workflow validation, packaged-consumer checks, and a publication dry run pass locally on macOS.
• CI covers Linux, macOS, Windows, minimum Rhai, and a fresh consumer of the packaged crate.