Add LinearModel and CompactLinearModel objects to the statistics package

The GNU Octave Statistics package currently lacks an object-oriented framework for linear regression, missing the deep statistical diagnostics and memory-efficient deployment options found in MATLAB. This project aims to implement the LinearModel and CompactLinearModel classes to cleanly separate heavy model development from lightweight predictions. It will also modernize the fitlm function to support table inputs and Wilkinson formulas, and introduce a highly efficient stepwiselm function for automated model selection. The classes will include essential methods, extensive BIST testing, and comprehensive documentation, greatly improving Octave's statistical capabilities and MATLAB compatibility.