fix(studio): lazy-import extract_model_size_b in llama_cpp to fix transformers 5.x version switch

PR #4769 (e4d14992) added a module-level `from utils.models import
extract_model_size_b` to llama_cpp.py.  Since core/inference/__init__.py
eagerly imports llama_cpp, the spawn subprocess now hits:

  core.inference.__init__  →  llama_cpp  →  utils.models  →
  model_config (line 8: `from transformers import AutoConfig`)

…before _activate_transformers_version() has a chance to prepend
.venv_t5/ to sys.path.  Python caches transformers 4.57.6 in
sys.modules, and no subsequent sys.path change can override it.

Fix: replace the eager import with a thin wrapper that defers the
import to first call.  This breaks the transitive chain so
transformers is not loaded until after version activation.

studio/backend/core/inference/llama_cpp.py

@@ -52,8 +52,15 @@ _REPROMPT_MAX_CHARS = 2000
 _SHARD_FULL_RE = re.compile(r"^(.*)-(\d{5})-of-(\d{5})\.gguf$")
 _SHARD_RE = re.compile(r"^(.*)-\d{5}-of-\d{5}\.gguf$")
 
-# Model size extraction (shared with routes/inference.py)
+# Model size extraction — lazy import to avoid pulling in transformers
-from utils.models import extract_model_size_b as _extract_model_size_b
+# at module level (utils.models.model_config imports AutoConfig).
+# This module is imported by core/inference/__init__.py, which runs
+# before the subprocess version-activation code, so an eager import
+# here would cache transformers 4.x in sys.modules and break the
+# 5.x version switch.
+def _extract_model_size_b(model_id: str):
+    from utils.models import extract_model_size_b
+    return extract_model_size_b(model_id)
 
 # ── Pre-compiled patterns for tool XML stripping ─────────────
 _TOOL_CLOSED_PATS = [