Add AIPW docstring warning to `get_toc()` for observational data

[aman-coder03]

Currently, the observed-outcome mode in get_toc() estimates the TOC via naive difference-in-means..

    subset_ATE = mean(Y | W=1, top-q) - mean(Y | W=0, top-q)

This is valid for RCT data but may be biased for observational data. As noted in Yadlowsky et al. (2021) Section 4, the correct approach for observational settings is to use doubly-robust AIPW pseudo-outcomes

The fix is a docstring note in get_toc() clarifying this limitation and suggesting users pass AIPW pseudo-outcomes as treatment_effect_col for observational data
related to #887