Recognize gemma3/llama4/mistral-small3.1+/multimodal as vision models (#1430)

is_vision_model() classified several genuinely multimodal families as text-only
because their names contain neither "vision" nor "vl": Gemma 3 (4b+), Llama 4,
Mistral Small 3.1/3.2, and *-multimodal models (e.g. phi-4-multimodal). For those
the attached image was stripped before the request, so the model never saw it —
a "can't read the image" report (issue #1274), common with Ollama tags like
gemma3:4b.

Add those keywords (plus a generic "multimodal"). Per the file's err-toward-True
policy (#124), a rare text-only tag treated as vision is the safer failure than
dropping a real image. Guard tests confirm the text-only siblings (gemma2, plain
gemma, mistral-small, phi-3) are not over-matched.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_vision_model_detection.py

@@ -28,3 +28,21 @@ def test_text_only_models_not_flagged():
 
 def test_none_is_safe():
     assert is_vision_model(None) is False
+
+
+def test_recognizes_multimodal_families_without_vision_in_name():
+    # issue #1274: these are vision-capable but their names don't contain
+    # "vision"/"vl", so they were dropped and the model never saw the image.
+    for name in [
+        "gemma3:4b", "gemma3", "gemma-3-27b-it",
+        "llama4:scout", "llama4", "llama-4-maverick",
+        "mistral-small3.1", "mistral-small-3.2",
+        "phi-4-multimodal", "phi4-multimodal",
+    ]:
+        assert is_vision_model(name), f"{name!r} should be detected as vision-capable"
+
+
+def test_new_keywords_do_not_overmatch_text_models():
+    # The added families must not flag their text-only siblings.
+    for name in ["gemma2:9b", "gemma:7b", "llama3.3", "mistral-small", "phi-3-mini"]:
+        assert not is_vision_model(name), f"{name!r} should not be flagged as vision"