Wrap external explainability tooling around the package's models

[breimanntools]

Problem

The parent epic names downstream explainability wrappers as a bridge, with no owner and nothing built. It is mentioned only in the decision layer and the ecosystem diagram.

Goal

Deliver this bridge as documented, tested work rather than an unowned line on an epic.

Requirements

• A wrapper letting external explainability tools consume the fitted models and feature matrix
• Optional dependency, gated the way the other heavy extras are, never in core
• Docstring, example notebook and unit tests
• A worked example on one dataset

KPIs / Acceptance criteria

• An external explainability tool produces attributions for a fitted model here
• A base install without the extra still imports cleanly

Scope / non-goals

• No heavy dependency in core. Anything heavy is an optional extra, gated like the existing ones.
• One bridge only. Sibling bridges have their own issues.

Child of #210.