Add real-model integration test pipeline

[baijumeswani]

Adds an ADO pipeline that validates the ORT GenAI native layer against real models pulled from the foundrylocalmodels storage account.

What it does

For every (model, execution provider) pair, the pipeline:

1. Builds an ORT GenAI wheel from source (one build per OS / arch).
2. Spawns a separate ADO test job that downloads only its model from blob, installs the wheel, and runs pytest.

Coverage

OS / arch	Providers tested
Windows x64	CPU, CUDA, WebGPU
Linux x64	CPU, CUDA
macOS arm64	CPU, WebGPU

Suites

The hardwired model catalog lives in test/python/integration/models.py:

• pr (6 small models, one per architecture family): runs on every PR.
• all (22 models, broader coverage): runs on every merge to main.

See test/python/integration/README.md for the local workflow and the decision table for when to add a new model to pr vs all.

[Copilot]

Pull request overview

Adds an Azure DevOps pipeline and accompanying Python integration test suite to validate the ONNX Runtime GenAI native runtime end-to-end against real models (pulled per-job from the foundrylocalmodels storage account), across a matrix of OS/arch and execution providers.

Changes:

• Introduces a real-model pytest integration suite (test/python/integration/) with model cataloging + model-path resolution.
• Adds helper tooling (tools/python/integration/suite_paths.py) to compute per-(model, EP) blob subpaths for targeted azcopy downloads.
• Adds a new multi-stage ADO pipeline (.pipelines/integration-tests.yml + templates) to build wheels and fan out one test job per (model, EP).

Reviewed changes

Copilot reviewed 13 out of 14 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tools/python/integration/suite_paths.py	CLI helper to print the blob subpath for a given (model, device) pair.
test/python/integration/test_integration_text.py	New text-generation integration test exercising model load + greedy generation for a chosen EP.
test/python/integration/resolver.py	Resolves logical model id + device to the newest on-disk vN/ model directory under ORTGENAI_MODEL_ROOT.
test/python/integration/README.md	Documents local + CI workflow, model layout, and suite membership guidelines.
test/python/integration/models.py	Defines the hardwired Foundry model catalog and the pr/all suites.
test/python/integration/conftest.py	Adds pytest CLI options and parametrization for --model, --execution-provider, and model root.
test/python/integration/init.py	Marks the integration directory as a Python package.
test/python/conftest.py	Makes --test_models optional so integration tests can run without providing unit-test model assets.
.pipelines/stages/jobs/steps/integration-pytest-step.yml	Template step to install the built wheel + deps and run the integration pytest suite.
.pipelines/stages/jobs/steps/integration-fetch-models-step.yml	Template step to azcopy a single model’s blob prefix onto the agent.
.pipelines/stages/jobs/integration-test-job.yml	Defines the per-(model, EP) test job (downloads wheel + model, runs pytest).
.pipelines/stages/jobs/integration-build-job.yml	Defines the per-OS/arch wheel build job used by the integration pipeline.
.pipelines/stages/integration-stage.yml	Orchestrates build stages and fans out per-model test jobs per OS/EP.
.pipelines/integration-tests.yml	Top-level ADO pipeline wiring: triggers, parameters, suite selection, and stage composition.