Workflow for executing examples

[oleksandr-pavlyk]

This PR

1. Fixes code in examples to change stream argument of MemoryResource.allocate method from position to keyword. Recent changes in cuda.core made the argument keyword-only one.

2. Adds python/examples/stream.py example, parallel to C++ examples/stream.cu

3. Adds a workflow to run examples periodically, manually and as part of PR to ensure that examples remain runnable.

   • Manual and periodic runs use latest versions of packages used by python/examples/*.py from main branch
   • PR jobs are constrained to versions known to have worked for source in main branch, and execute sources from the branch
   • Manual and periodic runs output versions used in the format suitable for updating the constraint file python/examples/requirements/constraints-cu13-py313.txt

Closes #444

[coderabbitai]

📝 Walkthrough

Summary by CodeRabbit

• New Features

  • Added a stream-processing benchmark example for measuring GPU memory and element performance.
  • Added support for running Python examples across CPU and GPU configurations, with selectable groups and clear results.
  • Added optional heavy-example and flexible-dependency test modes, including scheduled GPU validation.

• Bug Fixes

  • Improved CUDA stream handling in several Python examples for more reliable device-memory operations.

• Chores

  • Consolidated dependencies for CUDA 13 and Python 3.13 testing.
  • Added automated dependency consistency checks and expanded example test coverage.

Walkthrough

Summary

The pull request adds GPU CI coverage for Python examples, containerized test scripts, dependency constraints, an example runner, stream keyword fixes, and a CuPy benchmark.

Changes

Python example CI

Layer / File(s)	Summary
Example runner ci/run_python_examples.py	Adds configured example cases, group selection, dependency checks, subprocess execution, timeout handling, syntax checks, result reporting, and exit codes.
Containerized example environments ci/test_python_examples.sh, ci/test_python_examples_inner.sh	Runs grouped Python examples in a CUDA container with isolated environments and optional floating-dependency testing.
Example dependency definitions and constraint reporting ci/format_python_example_constraints.py, ci/test_format_python_example_constraints.py, python/examples/requirements/*	Adds recursive requirements, Python 3.13 CUDA 13 constraints, freeze collection, constraint candidate reporting, and parser tests.
Scheduled and gated workflow execution .github/workflows/build-and-test-python-wheels.yml	Adds configurable inputs, scheduled heavy and floating-dependency runs, GPU example testing, and workflow failure handling.
Example allocation and benchmark updates python/examples/axes.py, python/examples/cute_dsl_sgemm.py, python/examples/exec_tag_sync.py, python/examples/exec_tag_timer.py, python/examples/stream.py	Changes device allocations to pass stream by keyword, centralizes the default CuTe stream, and adds a CuPy stream benchmark.

Assessment against linked issues

Objective	Addressed	Explanation
Update Python examples to pass stream as a keyword argument [444]	✅
Add CI coverage that detects Python example breakages [444]	✅

Assessment against linked issues: Out-of-scope changes

Code Change	Explanation
New CuPy stream benchmark (python/examples/stream.py)	This adds a new benchmark rather than updating failing allocation calls or adding CI coverage for existing examples.

---

_{Comment @coderabbitai help to get the list of available commands.}

[shwina]

Python changes look good to me. I didn't review the CI scripts in much detail but they should be fine.