feat(kb): type ask/post results via binding_model keyword

[david-debest]

Closes #57

Problem

kb.ask() / kb.ask_with_info() (and the POST equivalents) are typed as returning Sequence[BindingModel] | BindingSet, even when the KI was registered with a binding model. At runtime the results are already validated into that model, but the type checker can't know it: the KI is looked up by a ki_name string. Callers had to use cast() or # type: ignore.

Solution

All outgoing methods (ask, ask_with_info, post, post_with_info and their *_sync bridges) take a keyword-only argument that types the result:

people = await kb.ask(bs, "ask-ki", binding_model=PersonBinding)                  # Sequence[PersonBinding]
res = await kb.post_with_info(bs, "post-ki", result_binding_model=ResultBinding)  # ExchangeResult[ResultBinding]

• binding_model (ASK) / result_binding_model (POST) mirror the parameter names of ask_ki() / post_ki().
• The argument only sets the result type: parsing still uses the model registered for the KI. Passing a different model (including one for a KI registered without a model) raises ValueError, so the static type can't lie.
• Uses PEP 696 TypeVar defaults (Python 3.13+), so each method has a single signature and no overloads.
• ExchangeResult is now generic in its binding type.
• Examples 03, 04 and 07 pass the model (07 drops its # type: ignore workarounds); README and CONTEXT.md are updated.

Behaviour change (typing only)

Without the keyword, the result type is now Sequence[BindingModel | dict[str, str]] instead of Sequence[BindingModel] | Sequence[dict[str, str]]. Runtime behaviour is unchanged, but code that annotated a variable with the old union may get a new type-checker error.

Verification

• New tests: typed results for ask/post, ValueError on a mismatching model (async and sync bridges).
• Checked inference with both pyright and mypy: Sequence[P] / ExchangeResult[Q] with a model, the union without one.

Notes

• The package ships no py.typed marker, so mypy users of the installed package won't see these types yet; see Ship a py.typed marker so type checkers use the package's inline types #63.