fix: capability probe promotes both backends independently

detect_default_types() previously fell back to {mlx} when mlx_lm
wasn't importable, which would cause charon (Linux, no mlx) to try
mlx targets and fail. Now each backend is probed independently:

mlx = mlx_lm importable + Darwin
gguf = openai importable

Studio (mac) sees {mlx, gguf} since openai is always in the PEP 723
closure. Charon (Linux) sees {gguf} since mlx_lm import raises. Empty
detection triggers an explicit error in main() asking the caller to
set LEM_TYPES, instead of silently routing to an empty target list.

Co-Authored-By: Virgil <virgil@lethean.io>

eval.py

@@ -748,22 +748,34 @@ SUPPORTED_TYPES = {"mlx", "gguf"}
 def detect_default_types():
     """Figure out which target types this machine can run by capability probe.
 
-    Apple Silicon + mlx_lm installed → mlx. Anything else (or explicit opt-in)
-    → gguf via an OpenAI-compatible endpoint (Ollama / llama.cpp server).
-    Returns a set. Workers override with --type or the LEM_TYPES env var.
+    Each type is probed independently and added if the dependency is present:
+
+      mlx  : Apple Silicon (Darwin) with mlx_lm importable
+      gguf : any machine with the `openai` client importable (Ollama's
+             endpoint availability is checked at wrapper __init__ time
+             via a lightweight probe request)
+
+    Workers override with --type <name> or the LEM_TYPES env var. If the
+    probe finds zero types, the caller gets an explicit error from main()
+    asking them to set LEM_TYPES.
     """
     import platform
     types = set()
+
     try:
         import mlx_lm  # noqa: F401
         if platform.system() == "Darwin":
             types.add("mlx")
     except ImportError:
         pass
-    # gguf path will be available once gguf_wrapper lands — for now it's
-    # opt-in via explicit --type gguf so workers don't silently skip
-    # mlx-only targets when the wrapper is absent.
-    return types or {"mlx"}
+
+    try:
+        import openai  # noqa: F401
+        types.add("gguf")
+    except ImportError:
+        pass
+
+    return types
 
 
 def _print_target_table(targets, highlight_types=None):
@@ -819,6 +831,12 @@ def main():
         allowed_types = set(os.environ["LEM_TYPES"].split(","))
     else:
         allowed_types = detect_default_types()
+        if not allowed_types:
+            parser.error(
+                "no target types detected on this machine (tried mlx, gguf). "
+                "Install mlx_lm (Apple Silicon) or openai (any), or set LEM_TYPES "
+                "explicitly in the environment."
+            )
 
     if args.list_targets:
         _print_target_table(all_targets, highlight_types=allowed_types)