fix: handle llama_cpp/llguidance version incompatibility gracefully

[octo-patch]

Fixes #1384

Problem

When llama_cpp is installed but at a version incompatible with llguidance.llamacpp
(e.g. llama-cpp-python==0.2.x with llguidance>=1.5.0), the module-level
import llguidance.llamacpp raises an AttributeError (e.g. module 'llama_cpp' has no attribute 'llama_vocab_p') that crashes the entire guidance import.

This makes the library completely unusable even for users who are not using LlamaCpp
at all, since guidance/models/__init__.py imports from _llama_cpp.py at module level.

Verified on: llama-cpp-python==0.2.7 + llguidance==1.5.0:

ImportError while loading conftest ...
guidance/models/_llama_cpp.py:25: in <module>
    import llguidance.llamacpp
AttributeError: module 'llama_cpp' has no attribute 'llama_vocab_p'

Solution

Consolidate both import llama_cpp and import llguidance.llamacpp into a single
try/except block that catches ModuleNotFoundError, ImportError, and
AttributeError. When either package is missing or incompatible, is_llama_cpp
is set to False and the rest of guidance loads normally.

Users who attempt to instantiate LlamaCpp will still receive a clear error message
at instantiation time (the existing check at the top of LlamaCppEngine.__init__).

Testing

• Verified import guidance succeeds with llama-cpp-python==0.2.7 installed
• Unit tests pass: pytest tests/unit/library/test_gen.py tests/unit/test_decorator.py
  (44 passed, 1 xfailed)