MLX - dynamic loading of mlx-c (#13735)

* MLX - dynamic loading of mlx-c

Create a wrapper layer to indirect the dependency on mlx-c so
the main ollama binary does not have a load-time dependency on mlx-c, mlx, and on linux, cuda.  Lazy load the library via dlopen
so we can adjust the path to ensure the dependencies are found
and fail gracefully if not present.

* review comments

* fix broken tests

x/imagegen/cmd/engine/generate.go

@@ -65,12 +65,12 @@ func (s *utf8Streamer) Flush() string {
 	return result
 }
 
-func init() {
-	generationStream = mlx.NewStream()
-}
-
 // withStream runs fn with the generation stream as default
 func withStream(fn func()) {
+	// Lazy initialization of generationStream
+	if generationStream == nil {
+		generationStream = mlx.NewStream()
+	}
 	orig := mlx.GetDefaultStream()
 	mlx.SetDefaultStream(generationStream)
 	fn()

x/imagegen/cmd/engine/main.go

@@ -12,7 +12,6 @@ import (
 	"path/filepath"
 	"runtime/pprof"
 
-	"github.com/ollama/ollama/x/imagegen"
 	"github.com/ollama/ollama/x/imagegen/mlx"
 	"github.com/ollama/ollama/x/imagegen/models/gemma3"
 	"github.com/ollama/ollama/x/imagegen/models/gpt_oss"
@@ -79,6 +78,11 @@ func main() {
 		return
 	}
 
+	// Check if MLX initialized successfully
+	if !mlx.IsMLXAvailable() {
+		log.Fatalf("MLX initialization failed: %v", mlx.GetMLXInitError())
+	}
+
 	// CPU profiling
 	if *cpuProfile != "" {
 		f, err := os.Create(*cpuProfile)