[BUG] Fix Pipeline._transform not chaining transformer outputs

[WHOIM1205]

Title

Fix Pipeline._transform to correctly chain transformer outputs during prediction

---

Summary

This PR fixes a critical bug in Pipeline._transform where transformer outputs were not chained during inference.

Previously, each transformer in the pipeline received the original, untransformed X, and only the last transformer's output (applied to raw input) was passed to the regressor.

This caused:

• Silent prediction corruption for pipelines with 2+ transformers
• Incorrect probabilistic outputs (predict_proba, predict_interval, predict_quantiles, predict_var)
• Invalid cross-validation and evaluation results
• UnboundLocalError for regressor-only pipelines (0 transformers)

The fix mirrors the chaining logic already used in _fit by reassigning X inside the loop.

---

Root Cause

In _transform, the output of each transformer was assigned to Xt, but X was never updated:

Xt = transformer.transform(X=X)

[WHOIM1205]

@fkiraly
fixes a bug in Pipeline._transform where transformer outputs were not chained during prediction
each transformer was incorrectly applied to the original X causing silent prediction errors for multi transformer pipelines the fix reassigns X inside the loop (matching _fit) so transformations are properly chained also resolves UnboundLocalError for regressor only pipelines

[fkiraly]

I think this is a duplicate of #780?

[WHOIM1205]

I think this is a duplicate of #780?

so should i close this ??