llm,server: route Ollama-format gemma3 blobs through llama/compat

Two tiny Go-side changes that let the llama/compat shim take over gemma3:

1. llm/llama_server.go: when the GGUF has embedded v.* tensors and no
   projector layer is declared, pass the model file itself as --mmproj.
   The in-process compat layer translates the same file into both a
   text-only view (for --model) and a clip-mmproj view (for --mmproj).

2. server/model_resolver.go: drop library/gemma3 from compatModelRedirects.
   The compat layer handles it directly, so no dhiltgen/ republish is
   needed. Other arches stay in the redirect list until they get their
   own handler in llama/compat/llama-ollama-compat.cpp.

End-to-end verified: `ollama run gemma3` answers text and image prompts
against the existing library/gemma3 blob with no re-download.

llm/llama_server.go

@@ -423,6 +423,15 @@ func NewLlamaServerRunner(
 	// Check if this is an embedding model
 	_, isEmbedding := f.KV()[fmt.Sprintf("%s.pooling_type", f.KV().Architecture())]
 
+	// Older Ollama-format GGUFs store vision tensors (v.*, mm.*) inline in
+	// the main model file rather than in a separate projector layer. Detect
+	// this case and point --mmproj at the model itself — the in-process
+	// llama.cpp compat shim translates the same file into both a text-only
+	// view and a clip-mmproj view. See llama/compat/ for details.
+	if len(projectors) == 0 && len(f.Tensors().Items("v.")) > 0 {
+		projectors = []string{modelPath}
+	}
+
 	gpuLibs := ml.LibraryPaths(gpus)
 	status := NewStatusWriter(os.Stderr)
 

server/model_resolver.go

@@ -8,10 +8,15 @@ import (
 	"github.com/ollama/ollama/types/model"
 )
 
-// Temporary redirection logic to map incompatible library models to compatible versions
+// Temporary redirection logic to map incompatible library models to compatible versions.
+//
+// Architectures listed here are handled via republished blobs under the
+// dhiltgen/ namespace. Once llama/compat/ grows a handler for an arch, its
+// entry should be removed from this list — the compat layer translates the
+// original library/ blob in memory so no republish is needed.
 var compatModelRedirects = []struct{ from, to string }{
 	{"library/gpt-oss", "dhiltgen/gpt-oss"},
-	{"library/gemma3", "dhiltgen/gemma3"},
+	// library/gemma3 — handled by llama/compat (text + vision).
 	{"library/embeddinggemma", "dhiltgen/embeddinggemma"},
 	{"library/snowflake-arctic-embed2", "dhiltgen/snowflake-arctic-embed2"},
 	{"library/gemma3n", "dhiltgen/gemma3n"},