fix: correct tooling config mismatches and add CI lint workflow

[KuaaMU]

Summary

This PR fixes the code quality tooling configuration mismatches and adds a CI workflow to enforce linting on every push/PR.

Problems Found

The project configured black, isort, and flake8 in pyproject.toml, but the tools were broken:

1. black couldn't parse files: target-version = ['py38'] but code uses match/case (Python 3.10+)
2. flake8 line-length mismatch: No config file → defaults to 79 chars while black uses 100
3. Actual formatting violations: Trailing whitespace, missing newlines, long lines
4. No CI: Tools configured but never enforced

Changes

Configuration Fixes

• pyproject.toml: target-version py38 → py310, requires-python >=3.8 → >=3.10
• .flake8: Add config with max-line-length = 100 to match black
• pyproject.toml: Remove Python 3.8/3.9 classifiers (match/case requires 3.10+)

Code Formatting

• Run black and isort on all Python files to fix existing violations

CI

• Add .github/workflows/lint.yml that runs black, isort, and flake8 checks

Housekeeping

• Add .gitignore for .venv/ and pycache/

Verification

All linters pass locally:

✅ black --check .
✅ isort --check .
✅ flake8 batch_invariant_ops/ test_batch_invariance.py deterministic_vllm_inference.py

Notes

• CI only runs linting (no tests) since the project requires CUDA/Triton for runtime testing
• This complements existing PRs (Add test suite for batch-invariant operations #16 test suite, Fix: Replace tl.range flatten with Python range for Triton 3.x compatibility #19 Triton compat) without overlapping
• The code changes are formatting-only (no functional changes)