Add performance metrics for StructuredOutput generation

[pavel-esir]

Ticket: CVS-168960

[Copilot]

Pull Request Overview

This PR instruments and exposes performance metrics for structured output generation, capturing grammar compilation and initialization timings across C++ and Python layers, along with tests and CI updates.

• Extend RawPerfMetrics and PerfMetrics to include grammar compile and compiler-init times, with new SummaryStats type.
• Propagate timings through StructuredOutputController, Sampler, and ContinuousBatchingPipeline in C++.
• Bind new metrics to Python, update type stubs, add a test for structured output metrics, and tweak CI to skip it in precommit.

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/python_tests/test_llm_pipeline.py	Added test_perf_metrics_with_structured_output test case
src/python/py_utils.cpp	Handle StructuredOutputConfig in kwargs-to-AnyMap conversion
src/python/py_perf_metrics.cpp	Bind and document new grammar compile/init metrics in Python
src/python/openvino_genai/py_openvino_genai.pyi	Updated stub to include SummaryStats and new PerfMetrics methods
src/cpp/src/sampling/structured_output/structured_output_controller.hpp	Record and expose init/compile times, added std::mutex field
src/cpp/src/sampling/structured_output/structured_output_controller.cpp	Time grammar compilation and initialization
src/cpp/src/sampling/sampler.hpp	Expose get_structured_output_times and clear method
src/cpp/src/sampling/sampler.cpp	Implement structured-output timing getters
src/cpp/src/perf_metrics.cpp	Implement calc_full_stat, bind new metrics, merge operations
src/cpp/src/generation_config.cpp	Read structured_output_config from AnyMap
src/cpp/src/continuous_batching/pipeline_impl.cpp	Clear and record structured-output metrics in batch pipeline
src/cpp/include/openvino/genai/perf_metrics.hpp	Extend RawPerfMetrics/PerfMetrics definitions
.github/workflows/linux.yml	Skip new structured-output test via -k filter

Comments suppressed due to low confidence (2)

.github/workflows/linux.yml:906

• The -k "not test_perf_metrics_with_structured_output" filter prevents the new structured-output performance test from running in CI. Consider removing or adjusting this filter so the test executes under the appropriate job (e.g., nightly).

          python3 -m pytest -v ${{ env.SRC_DIR }}/tests/python_tests/test_llm_pipeline.py -k "not test_perf_metrics_with_structured_output"

src/cpp/src/sampling/structured_output/structured_output_controller.hpp:87

• The code uses std::mutex but the header <mutex> is not included in this file, which will cause a compilation error. Please add #include <mutex> at the top.

    std::mutex m_mutex;

[Wovchena]

There's merge conflict

[pavel-esir]

There's merge conflict

resolved