feat(skill): consider supporting InferenceFormatFactory in SkillGenerator

### Background

Currently in PR #2516, `SkillGenerator` wraps an instantiated `InferenceFormat` strategy instance:

```python
generator = SkillGenerator(fmt)
skill = generator.generate_skill() # defaults to catalogs in fmt
```

In the Agent SDK module blueprint (`blueprints/modules/a2ui_agent.blueprint.md` §4.E), format construction is planned to be abstracted through `InferenceFormatFactory` for the upcoming `a2ui.processor` facade:

```python
factory: InferenceFormatFactory
fmt = factory.create(catalog, capabilities)
```

### Proposed Enhancement

Once `InferenceFormatFactory` is implemented in the Python Agent SDK and multi-language core SDKs, evaluate whether `SkillGenerator` should accept an `InferenceFormatFactory` directly, allowing catalogs and format options to be synthesized on the fly without pre-bound format instances.

Review the SkillGenerator class and the InferenceFormat interface in the codebase, likely in the Python Agent SDK. Examine the blueprint at blueprints/modules/a2ui_agent.blueprint.md section 4.E to understand the planned InferenceFormatFactory. Determine how catalogs and capabilities are currently passed and how a factory would change the flow. Check for existing tests of SkillGenerator to understand the expected behavior.

Rédigé par le modèle d'indexation à partir du texte de l'issue.