Add bootstrap inference options to LDTE/LPTE (partial compliance)

[TomeHirata]

Follow-up to #150 for the optional point in the JOSS review (openjournals/joss-reviews#11047): the partial-compliance methods didn't expose bootstrap inference options, unlike predict_dte/predict_pte.

Changes

• predict_ldte / predict_lpte on SimpleLocalDistributionEstimator and AdjustedLocalDistributionEstimator (and the public compute_ldte / compute_lpte) take variance_type and n_bootstrap:
  • "moment" (default): the existing analytic pointwise intervals. Results are unchanged.
  • "multiplier": pointwise intervals from a multiplier bootstrap.
  • "uniform": a band that holds over all locations at once, using a max-t critical value from the multiplier bootstrap. Locations with zero variance (e.g. a CDF at or above the max outcome) are left out of the max.
  • Any other value raises ValueError.
• How the bootstrap works: it reuses the per-observation influence functions the analytic variance already computes, so no models are refit. The stratum×arm terms are centered within each cell and the stratum term is constant within a stratum, so the cross terms cancel and mean(influence**2) equals the existing analytic sigma exactly. That makes "multiplier" a bootstrap of the same estimator as "moment". It uses the same multipliers as compute_confidence_intervals.
• Argument order: variance_type and n_bootstrap sit before display_progress, matching predict_dte. Calls that pass display_progress positionally would need updating. All in-repo calls use keywords.
• Docs: the Oregon tutorial gets a short note on the new options.

Testing

• New tests:
  • the default equals explicit "moment"
  • "multiplier" half-widths are within 10% of "moment"
  • the "uniform" band is wider than the pointwise one and contains the estimate
  • LPTE with the adjusted estimator works for both bootstrap types
  • an invalid variance_type raises
  • unit tests for the bootstrap helper, covering its standard error and how zero-variance locations are handled
• Full suite: 79 passed. ruff check . is clean.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.