Refactor trans_models_t to use mlr3: replace fit_fun/gof_fun with Learner/AutoTuner interface

[Copilot]

Replaces the ad-hoc fit_fun/gof_fun function-passing interface with first-class mlr3 integration. Learner identity, hyperparameters, and a serialized untrained spec are stored natively in DuckDB; cross-validation uses mlr3 tasks and measures throughout.

Schema (trans_models_t)

Old	New	Type	Notes
model_family	learner_id	VARCHAR	mlr3 learner key
model_params	learner_params	MAP(VARCHAR,VARCHAR)	Atomic scalar params only
fit_call	learner_spec	BLOB	Serialized untrained Learner
goodness_of_fit	crossval_measures	MAP(VARCHAR,DOUBLE)	prediction$score(measures)
model_obj_part	crossval_predictions	BLOB	Serialized PredictionClassif
model_obj_full	learner_full	BLOB	Serialized trained Learner

Primary key: (id_run, id_trans, fit_call) → (id_run, id_trans, learner_id)

API

# New signatures — no backward compatibility
fit_partial_models(self, learner, measures, sample_frac = 0.7, seed = NULL, cluster = NULL)
fit_full_models(self, learner, measures, gof_criterion, gof_maximize, cluster = NULL)

# Example
db$trans_models_t <- db$fit_partial_models(
  learner  = mlr3::lrn("classif.ranger", num.trees = 500, predict_type = "prob"),
  measures = list(mlr3::msr("classif.auc")),
  seed     = 42
)
db$trans_models_t <- db$fit_full_models(
  learner       = mlr3::lrn("classif.ranger", predict_type = "prob"),
  measures      = list(mlr3::msr("classif.auc")),
  gof_criterion = "classif.auc",
  gof_maximize  = TRUE
)

Worker logic

• fit_partial_model_worker: Builds as_task_classif(..., positive = "TRUE"), deep-clones and trains the learner, scores held-out split via prediction$score(measures). For AutoTuner, extracts the optimal inner learner for learner_id/learner_params/learner_spec.
• fit_full_model_worker: Reconstructs from learner_spec BLOB; falls back to do.call(mlr3::lrn, c(list(learner_id), as.list(learner_params))) on deserialization failure.
• predict_trans_pot: Deserializes learner_full and calls learner$predict_newdata(pred_data)$prob[, "TRUE"]; removes family-specific dispatch.

New method

get_crossval_plots(id_run, id_trans) deserializes all crossval_predictions BLOBs and returns mlr3viz::autoplot() results for visual GoF inspection.

Dependencies

mlr3 and mlr3viz added to Suggests.