Refactor perf regression runner to pytest with in-process execution

[Copilot]

Performance regression runs were executed via per-file subprocesses, providing poor progress visibility and brittle output parsing.

• Pytest-backed runner: regression_all now builds a transient pytest wrapper to run all regression benchmarks in-process with -s, keeping output visible and enabling richer reporting.
• Deterministic module loading: Bench files load via hashed, path-aware module names and safe cwd/sys.path context managers, with results aggregated from shared state.
• Configurable invocation: Optional pytest_args passthrough for custom flags; clearer error if pytest is missing.
• Tests: Added unit tests that stub tilelang.testing and pytest.main, validating bench execution and wrapper generation.

Example usage:

import tilelang.testing.perf_regression as pr

# Run all discovered regression benchmarks under examples/
pr.regression_all(pytest_args=["-q"])

Original prompt

This branch plans to do performance regression test between two commits, because when a developer give some pull request, we want to test these changes can lead to regression. Please use more robust to do this. Now, we do tests for each file in a subprocess. we want to use pytest etc. framework, so it can show process of tests. Refactor it.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.