CVS-175737-[OVEP] Expose kvcache_rewind python api

[Kotomi-Du]

Description

In Phi-Silica app which didn't rely on ORT-GenAI, we need an API to remove KV history. The kvcache_rewind is an OVEP function to achieve this, however there is no python API exposed. This PR is for this purpose.

If feature goes to new ABI?

Yes

Jira Ticket :

https://jira.devtools.intel.com/browse/CVS-175737

[ankitm3k]

Please attach a JIRA for this feature request in the PR description.

[Kotomi-Du]

Please attach a JIRA for this feature request in the PR description.

done

[Copilot]

Pull Request Overview

This PR exposes a Python API for the kvcache_rewind functionality from OVEP (OpenVINO Execution Provider), enabling applications like Phi-Silica to manage KV cache history without relying on ORT-GenAI. The implementation adds a generic set_ep_dynamic_options method that passes dynamic configuration options to execution providers at runtime.

Key Changes:

• Added set_ep_dynamic_options method to enable runtime configuration of execution providers
• Implemented Python bindings in both the C++ pybind layer and Python wrapper class
• Provided comprehensive documentation with usage examples

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
onnxruntime/python/onnxruntime_pybind_state.cc	Implements C++ pybind11 binding for set_ep_dynamic_options with dict-to-C-array conversion and error handling
onnxruntime/python/onnxruntime_inference_collection.py	Adds Python wrapper method with type hints and documentation

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[MayureshV1]

@preetha-intel - Can you please review from perspective of setting workload type using this python interface.
@ankitm3k , @RyanMetcalfeInt8 - Please review whether the interface can be used to set kvcache rewind using Python.

[MayureshV1]

LGTM !

[preetha-intel]

LGTM

[RyanMetcalfeInt8]

LGTM!