chore: FSD stabilization, strict TDD enforcement, and unlearn mechanism

- implemented self-healing unlearn for Qdrant false positives
- centralized testing logic in conftest
- documented core rules, ai standards, and goap philosophy
- purged old dev scratchpads

tests/e2e/conftest.py

@@ -251,6 +251,21 @@ def e2e_configs():
         ai_telepathic_url="http://localhost",
         ai_telepathic_model="llama3",
         ai_condenser_url="http://localhost",
-        ai_condenser_model="llama3"
     )
     return configs
+
+@pytest.fixture(autouse=True)
+def mock_sae_perceive(request, monkeypatch):
+    """
+    Mock SAE.perceive for all E2E tests EXCEPT the ones actually testing SAE.
+    This prevents the tests from hitting the local Qdrant/Ollama instances
+    and failing due to non-deterministic LLM output or missing caches.
+    """
+    if "test_e2e_sae.py" in str(request.node.fspath):
+        return
+    if request.config.getoption("--live"):
+        return
+        
+    import GramAddict.core.situational_awareness
+    monkeypatch.setattr(GramAddict.core.situational_awareness.SituationalAwarenessEngine, "perceive", lambda self, xml: GramAddict.core.situational_awareness.SituationType.NORMAL)
+