[BUG] Support dynamic exogenous features in transfer intervals

[janrth]

Summary

Fixes #721.

Transfer conformal prediction could not handle dynamic exogenous inputs consistently:

• recalibrate and error_scaled run a frozen target-domain backtest, but its nested predictions did not receive the validation fold's future dynamic inputs.
• weighted_conformal and scale_aligned_weighted preprocess target history for density-ratio estimation, but did not preserve the source model's fitted static/dynamic feature configuration.

This PR keeps the fix local to transfer-conformal internals:

• _frozen_backtest resolves dynamic model-input columns from the fitted feature schema once, then passes each validation fold's values as X_df to its nested prediction.
• The weighted-transfer preprocessing callback preserves the fitted id_col, time_col, target_col, and static_features configuration.

Ordinary predict() and ordinary cross_validation() behavior are unchanged.

Behavior

All five transfer methods now work with dynamic exogenous features:

• recalibrate
• error_scaled
• scale_aligned
• weighted_conformal
• scale_aligned_weighted

The added regression test uses a deterministic linear system with a dynamic input and verifies that every method creates interval columns while leaving point forecasts identical to the plain forecast.

[codspeed-hq]

Merging this PR will not alter performance

✅ 12 untouched benchmarks

---

_{Comparing janrth:bug/interval_transfer (a855643) with main (eff8ddb)}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: aab2149ae3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".