[PyOV] Fix passing Model as argument (openvinotoolkit#28896)

### Details:
- `run_on_model` was broken due-to changes introduced by this PR
openvinotoolkit#27191
 - add more tests to cover such cases

### Tickets:
 - CVS-162131

Author: akuporos

src/bindings/python/src/pyopenvino/graph/passes/model_pass.cpp

@@ -10,6 +10,7 @@
 #include <string>
 
 #include "pyopenvino/core/common.hpp"
+#include "pyopenvino/utils/utils.hpp"
 
 namespace py = pybind11;
 
@@ -34,10 +35,14 @@ void regclass_passes_ModelPass(py::module m) {
         "ModelPass");
     model_pass.doc() = "openvino.passes.ModelPass wraps ov::pass::ModelPass";
     model_pass.def(py::init<>());
-    model_pass.def("run_on_model",
-                   &ov::pass::ModelPass::run_on_model,
-                   py::arg("model"),
-                   R"(
+    model_pass.def(
+        "run_on_model",
+        [](ov::pass::ModelPass& self, const py::object& ie_api_model) {
+            const auto model = Common::utils::convert_to_model(ie_api_model);
+            self.run_on_model(model);
+        },
+        py::arg("model"),
+        R"(
                    run_on_model must be defined in inherited class. This method is used to work with Model directly.
 
                    :param model: openvino.Model to be transformed.

src/bindings/python/tests/test_transformations/test_model_pass.py

@@ -12,3 +12,10 @@ def test_model_pass():
     manager.run_passes(get_relu_model())
 
     assert model_pass.model_changed
+
+
+def test_model_pass_run_on_model():
+    model_pass = MyModelPass()
+    model_pass.run_on_model(get_relu_model())
+
+    assert model_pass.model_changed

src/bindings/python/tests/test_transformations/test_public_transformations.py

@@ -153,3 +153,14 @@ def test_serialize_pass(request, tmp_path, is_path_xml, is_path_bin):
 
     os.remove(xml_path)
     os.remove(bin_path)
+
+
+@pytest.mark.parametrize(("transformation", "arguments"),
+                         [(ConstantFolding, []),
+                          (MakeStateful, [{"parameter": "result"}]),
+                          (ConvertFP32ToFP16, []),
+                          (LowLatency2, [])])
+def test_run_on_model_transformations(transformation, arguments):
+    model = get_model()
+    transformation(*arguments).run_on_model(model)
+    assert model is not None