fix: avoid shadowing Python stdlib profile module

[renflowerz]

Summary

• prevent the repository's profile.py CLI from breaking Python's stdlib cProfile import
• expose the stdlib profile API only when the file is imported as the module named profile
• keep the existing AutoKernel CLI path unchanged when profile.py is executed
• add regression tests for both the stdlib API and a real cProfile.run() call from the repository root

Problem

From the repository root, Python resolves import profile to AutoKernel's profile.py instead of the standard-library module. cProfile imports profile by name and expects _Utils, run, and runctx, so the shadowed import fails while Torch/Transformers imports are initializing.

Before:

profile resolves to: .../autokernel/profile.py
AttributeError: module 'profile' has no attribute 'run'

After:

import cProfile; cProfile.run('sum(range(10))')
# succeeds from the repository root

The normal CLI still imports Torch and executes the existing profiler implementation when __name__ == "__main__".

Tests

• python3 -m unittest discover -s tests -v — Python 3.14, pass
• .venv/bin/python -m unittest discover -s tests -v — Python 3.10, pass
• 0652 B200 project venv:
  • compatibility tests pass
  • from transformers import AutoModelForCausalLM succeeds from repository root
  • uv run profile.py --help succeeds
  • real LLaMA profile smoke succeeds and writes a 20 KB JSON report
• python3 -m py_compile profile.py tests/test_profile_module_compat.py
• git diff --check