ENH: overlap diagnostics and explicit trimming for TreatmentEffect

Is your feature request related to a problem? Please describe

TreatmentEffect gives no way to check whether the overlap assumption is plausible to trust IPW-based results. When propensity scores get extreme, the weights explode and the estimates become unreliable. Currently, nothing in the API surfaces this and the user has to know to inspect prob_select themselves.

This came up while working on #10222: after making the propensity score clipping consistent, the estimate in that repro is internally consistent but still far from the true effect, because clipping only caps the extreme weights and doesn't tell the user that overlap is violated or let them restrict the sample to the region where it holds.

Describe the solution you'd like

Roughly what Stata provides with teoverlap and tebalance:

1. Overlap diagnostic: a summary of the propensity score distribution by treatment group, min/max, counts outside the clipping bounds, optionally a density plot, enough to see at a glance whether the groups share common support.
2. Balance table: standardized mean differences of the covariates before and after weighting, to check whether the propensity model actually balances the groups.
3. Explicit trimming: an option to drop observations with propensity scores outside given bounds, as an alternative to clipping. Unlike clipping this changes the estimand (effect on the overlap subpopulation), so it should be opt-in, documented as such, and report how many observations were dropped.

I'd be happy to work on this, starting with the diagnostics (1 and 2), which are purely additive.

Describe alternatives you have considered

No response

References

References

• Stata teoverlap (overlap plots): https://www.stata.com/manuals/causalteoverlap.pdf
• Stata tebalance (covariate balance diagnostics): https://www.stata.com/manuals/causaltebalance.pdf
• Crump, Hotz, Imbens, Mitnik (2009), "Dealing with limited overlap in estimation of average treatment effects", Biometrika 96(1)
• Austin & Stuart (2015), "Moving towards best practice when using inverse probability of treatment weighting", Statistics in Medicine 34(28), 3661–3679. https://doi.org/10.1002/sim.6607

Additional context

No response