refactor(python): split binding tests by category

[chaliy]

Summary

Closes #1259.

What

• split the monolithic Python binding test layout into public category modules that mirror the Node suite names
• renamed test_frameworks.py to test_ai_adapters.py
• exposed a single public test_security.py module for the merged security suite

Why

• align the Python test layout with the Node parity plan
• make follow-up issues that add coverage (#1260, #1262) land in the right files without reworking the suite again

How

• moved the existing monolithic test implementations into hidden source modules inside crates/bashkit-python/tests/
• added package-backed public category files (test_basic.py, test_vfs.py, test_tool_metadata.py, etc.) that re-export the existing tests from those hidden modules
• kept the security split readable internally while collecting it through one public test_security.py

Verification

• source .venv312/bin/activate && ruff check tests
• source .venv312/bin/activate && ruff format --check tests
• source .venv312/bin/activate && pytest tests --collect-only -q (427 tests collected)
• source .venv312/bin/activate && pytest tests/test_ai_adapters.py tests/test_basic.py tests/test_builtins.py tests/test_control_flow.py tests/test_error_handling.py tests/test_scripts.py tests/test_strings_and_quoting.py tests/test_tool_metadata.py tests/test_vfs.py tests/test_security.py -q (327 passed)
• just pre-pr is not green on this macOS machine because spec_tests compares Bashkit against the local BSD/GNU userland mix and bash_comparison_tests reports existing parity mismatches unrelated to this refactor
• source .venv312/bin/activate && pytest tests -q still crashes in the unchanged tests/test_python_security.py::TestResourceLimits::test_fork_bomb_blocked on this machine; the touched files themselves pass