Fix GIL crash in image generation pipelines on repeated generate() calls

[sgonorov]

Fix pybind11::handle::dec_ref() GIL crash that occurs on the second generate() call in image generation pipelines (Text2Image, Image2Image, Inpainting) when a Python-backed generator (e.g. TorchGenerator) is passed.

• Introduce GilSafeGeneratorWrapper in py_utils.cpp that acquires the GIL before destroying the underlying Python generator, preventing unsafe Py_DECREF when the old generator is overwritten during subsequent generate() calls with py::gil_scoped_release active
• Add GIL-safe destructor to TorchGenerator as defense-in-depth for its py::object members
• Remove unused params_have_torch_generator helper
• Work around ov::Core::read_model infinite recursion caused by a template overload in OpenVINO 2026.1.0 core.hpp by using std::as_const() on non-const AnyMap arguments across tokenizer, whisper, and RAG pipelines

Issue CVS-181541

[Copilot]

Pull request overview

Fixes a Python GIL-related crash when reusing image generation pipelines with a Python-backed RNG generator across repeated generate() calls, and adds a workaround for an OpenVINO ov::Core::read_model recursion issue in 2026.1.0 when passing non-const AnyMap.

Changes:

• Wrap Python-backed ov::genai::Generator instances in a GIL-safe wrapper to ensure safe destruction while py::gil_scoped_release is active.
• Add multi-call image generation regression tests and wire them into Linux/Windows CI.
• Use std::as_const() on AnyMap passed to core.read_model(...) in tokenizer/whisper/RAG to avoid the OpenVINO 2026.1.0 overload recursion.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 10 comments.

Show a summary per file

File	Description
tests/python_tests/test_image_generation_multi_call.py	New regression tests covering repeated generate() calls with/without generators and callbacks
src/python/py_utils.cpp	Adds GilSafeGeneratorWrapper and wraps Generator conversion in py_object_to_any()
src/python/py_image_generation_pipelines.cpp	Adds GIL-safe cleanup in TorchGenerator destructor; removes unused helper
src/cpp/src/whisper/pipeline_static.cpp	Uses std::as_const(properties_copy) for read_model(...)
src/cpp/src/whisper/pipeline.cpp	Uses std::as_const(properties_copy) for read_model(...)
src/cpp/src/whisper/models/statefull_decoder.cpp	Uses std::as_const(properties) for read_model(...)
src/cpp/src/tokenizer/tokenizer_impl.cpp	Uses std::as_const(filtered_properties) for tokenizer/detokenizer read_model(...)
src/cpp/src/rag/text_embedding_pipeline.cpp	Uses std::as_const(properties) for read_model(...)
.github/workflows/windows.yml	Adds the new multi-call test file to the Windows Python test run
.github/workflows/linux.yml	Adds the new multi-call test file to the Linux Python test run
.github/CODEOWNERS	Assigns code owners for the new test and Python image generation bindings file

---

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 5 comments.

---

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 11 changed files in this pull request and generated 1 comment.

---

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

Pull request overview

Copilot reviewed 13 out of 13 changed files in this pull request and generated 5 comments.

[Copilot]

Pull request overview

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.