[RFC 0005] Covariate Preprocessor Serialization and Standardization

[andrewherren]

• Keep R and Python native preprocessor schemas as-is (no cross-platform unification); they remain same-platform-only.
• Add cross_platform_portable boolean to the covariate_preprocessor object, computed at serialization time as (all covariates numeric).
• When cross_platform_portable == false, record the list of offending (non-numeric) columns.
• Cross-platform loader gate: if writer platform ≠ loader platform and flag is false → hard error naming offending column(s). Same-platform loads ignore the flag entirely.
• Numeric-only model: confirm the transform is identity and load/predict works in the other language.