Refactor common into cpu specific and truly common (#1817)

* refactor

* fix extension example

* fix no-cpu

mlx/backend/cpu/gemm.h

@@ -0,0 +1,20 @@
+// Copyright © 2025 Apple Inc.
+
+#pragma once
+#include "mlx/array.h"
+
+namespace mlx::core {
+
+template <typename T>
+void matmul(
+    const array& a,
+    const array& b,
+    array& out,
+    bool a_transposed,
+    bool b_transposed,
+    size_t lda,
+    size_t ldb,
+    float alpha,
+    float beta);
+
+} // namespace mlx::core