Support simulation from scratch when model has no dataset

[roninsightrx]

Summary

• create_sim_dataset(): when model$dataset is NULL (i.e. the $DATA file can't be found) and no data argument is supplied, build the simulation dataset from scratch if regimen is provided — otherwise error with a clear message. Uses a placeholder-marker approach to carry covariate values into dose rows via fill() without leaving ghost rows in the output.
• create_sim_dataset(): fix covariate column retention — when sim_data has no pre-existing covariate columns (from-scratch case), union(sim_data_cols, all_cov_cols) ensures newly-joined columns survive the select().
• run_sim(): abort with a clear message when model$dataset is NULL and data = NULL.
• run_sim(): validate required columns via get_required_input_variables() before running, giving an actionable error (naming the missing columns) when the dataset is incomplete.

Test plan

• create_sim_dataset(no-data): error when no dataset and no regimen
• create_sim_dataset(no-data): regimen-only produces correct dose + obs structure
• create_sim_dataset(no-data): n_subjects defaults to 1, or from nrow(covariates)
• create_sim_dataset(no-data): covariates are applied correctly and appear in output
• create_sim_dataset(no-data): no .placeholder column leaks into output
• run_sim(): data = NULL falls back to model$dataset (existing model with real data)
• run_sim(): clear error when model$dataset is NULL and data = NULL
• run_sim(): clear error naming the missing column when required column absent from dataset
• Existing tests unaffected (from-scratch path only triggers when model$dataset is NULL)

🤖 Generated with Claude Code