chore(lint): reduce Ruff exceptions

[tetsuo-cpp]

Summary

• add type annotations and explicit unused callback parameters to the GPT-2 demo
• infer attention launch dimensions from the input tensor to simplify the demo API
• rewrite GPU header parsing to comply with Ruff and remove avoidable per-file ignores
• retain only intentional exceptions for tests, subprocess tools, and CLI scripts

Verification

• uv run ruff check gpu_test demo
• uv run ruff format --check gpu_test demo
• uv run pytest -q gpu_test/test_compiler.py gpu_test/test_vast_session.py (10 passed)