fix: add chat_template property to Model for setting chat templates

[octo-patch]

Fixes #1196

Problem

The chat template smoke tests in tests/need_credentials/test_chat_templates.py were disabled with @pytest.mark.skip because the tests set lm.chat_template = ... on Mock model objects, but the Model base class had no chat_template property. As a result, setting lm.chat_template only set a plain attribute on the model instance, rather than updating the interpreter's chat template that is actually used for role-based formatting.

Solution

Add a chat_template property (getter and setter) to the Model base class in guidance/models/_base/_model.py:

• Getter: returns self._interpreter.chat_template if the interpreter supports it (via getattr with None as default for interpreters that do not have this attribute)
• Setter: sets self._interpreter.chat_template = value, which is then preserved across deepcopy when the model is copied during __add__ operations

This allows tests (and users) to configure the chat template on a model object directly:

lm = guidance.models.Mock("")
lm.chat_template = CHAT_TEMPLATE_CACHE[model_chat_template]()
# Now role blocks use the specified chat template
with guidance.user():
    lm += "Hello!"

Re-enable the two previously-skipped smoke tests: test_chat_format_smoke and test_chat_format_smoke_with_system.

Testing

The tests require HF_TOKEN and download actual HuggingFace models, so they run in the need_credentials suite. The fix is mechanically verified: lm.chat_template = value now properly sets lm._interpreter.chat_template = value, and subsequent deepcopy of the interpreter (triggered by model copy during __add__) preserves the new template.