docs: improve test suite documentation

[man4ish]

Summary

• Adds a module docstring and Developer: Manish Kumar <manish@omnibioai.org> attribution to all 14 test modules under tests/.
• Adds a docstring to all 107 test_* functions (98 were undocumented), each stating the specific behavior or security invariant the test asserts. Documents the fixtures. Existing docstrings are preserved verbatim.
• Documentation only: no imports, assertions, fixtures, markers, parametrization, mocks or application code changed. +305/−1 lines; the 1 removed line is a module docstring rewritten only to add the attribution line.

Verification

• AST equivalence: for every modified file, HEAD vs. working tree with docstring-only expression nodes stripped gives 0 executable differences (14 files).
• Syntax: 0 failures across 14 files.
• Focused tests (cache, security boundaries, tenancy, RBAC, engine permission/tenancy): 49 passed, 2 xfailed, identical before and after.
• Full suite: 128 passed, 2 xfailed, identical before and after.
• git diff --check is clean and only files under tests/ changed.

Notes for reviewers

• Like omnibioai-hpc-policy-engine, this repo ships Cython-compiled .so files next to their .py sources; Redis is mocked throughout the test suite, so all runs used a network-blocking pytest plugin as an extra check — 0 blocked connection attempts.
• Two tests remain xfail(strict=True) and are documented as such rather than as passing behavior: test_unknown_action_should_fail_closed and test_wildcard_action_should_not_bypass_authorization track known fail-open gaps in the production policy engine.
• tests/test_role_hierarchy_fixtures.py maps a conceptual Platform Owner → Org Admin → Scientist → Viewer hierarchy onto this engine's real role/permission vocabulary; that framing is preserved in the docstrings since it's load-bearing context from the module's own docstring, not decoration.

🤖 Generated with Claude Code