Preprocessing

q.preprocess owns transformations whose state is fitted at the model-training boundary. The namespace is intentionally separate from q.feature: feature values may be materialized for all dates, while fitted preprocessing must learn only from the training partition.

Namespace	Role
q.preprocess.impute	Missing-value indicators and imputation
q.preprocess.scale	Numeric normalization
q.preprocess.outlier	Outlier detection and treatment
q.preprocess.encode	Categorical representation
q.preprocess.reduction	Learned projection and decomposition
q.preprocess.selection	Selection of model-input columns

The packages are currently placeholders. Implementations will use a shared fit/transform contract, preserve pandas labels, and expose fitted artifacts for reuse in validation, testing, and online inference.