Categorical Treatments for Metalearners Disabled, But Docs Imply Metaleaners Handle Categorical Treatments

There seems to be some inconsistency between the documentation and implementation of metalearners. Docs state that "all of these estimators are applicable to multiple treatment settings as well" and the detailed estimator comparison table lists treatment type for metalearners as "categorical"; however, docstrings for the T argument in the fit() method for metalearners state "only binary treatments are accepted as input." Moreover, the check_inputs() utility in the fit methods hard-code multi_output_T to False. The metalearners appear to handle categorical treatments fine after simply removing the check_inputs() step (it could also be moved to after the one-hot-encoding).

Start with the metalearners' fit() docstrings and their check_inputs() calls, then compare them with the estimator comparison table's treatment-type claims. Determine whether validation should allow categorical treatments before or after one-hot encoding; done means the implementation, documentation, and input checks consistently support the intended treatment types.