Soften Python version check with general.yaml bypass

[Jammy2211]

Summary

• Replaces the hard RuntimeError for Python < 3.12 with a softer check that can be bypassed via version.python_version_check: False in a workspace's config/general.yaml.
• Updates the error message to accurately describe reality: Python 3.9/3.10/3.11 technically work but aren't officially tested against; 3.12+ is the supported track.
• Default behaviour on workspaces that haven't opted in is unchanged — the check still raises if Python < 3.12, just with a clearer message that names the working-but-unsupported versions and shows the YAML snippet to bypass.

Why

The PyAutoGPU venv is still on Python 3.10 and was being blocked from importing autoconf entirely, even though nothing in the library actually needs 3.12. The 3.12 cap was a support statement, not a functional dependency. Hard-failing at import is too aggressive — users with known-good older-Python environments should be able to opt in per-workspace.

API Changes

• autoconf/__init__.py: new private _python_version_check_bypassed() helper that does a minimal yaml.safe_load(Path.cwd() / "config" / "general.yaml") and returns True iff the version.python_version_check key is explicitly False. Any read/parse failure (missing file, missing yaml module, missing key, bad YAML) falls through to the raise, so the default fails safely.
• The raise now includes the YAML snippet the user needs to add to bypass.

Paired PR

• autolens_workspace: feature/soften-python-version-check adds the new version: block to the shipped config/general.yaml with python_version_check: True as the default.

Test plan

• pytest test_autoconf/ on Python 3.12 (87 passed)
• On PyAutoGPU (Python 3.10): python -c "import autoconf" with the default True setting → raises with the new multi-line message referencing 3.9/3.10/3.11 and the bypass snippet
• On PyAutoGPU (Python 3.10): flip to python_version_check: False → import succeeds, scripts/imaging/modeling.py (with PYAUTO_TEST_MODE=2) runs past the version gate into model-fitting
• cd /tmp && python -c "import autoconf" on 3.10 (no workspace config) → still raises, message unchanged

🤖 Generated with Claude Code