fix: eliminate thinking-block poisoning + no-op navigation trap

ROOT CAUSE: qwen3.5 (reasoning model) returns response='' with thinking
block containing all reasoning. llm_provider.py line 352 silently
substituted the thinking block as the response via:
  content = raw_response or raw_thinking or ''
The Brain then extracted random actions from the reasoning text.

FIXES:
1. llm_provider.py: Conditional thinking isolation
   - format_json=True (SAE/perception): thinking fallback preserved
   - format_json=False (Brain): thinking NEVER substituted
   - Added think=false for Ollama free-text calls to force direct response

2. planner.py: No-Op Guard strips tab actions that navigate to
   the current screen (e.g. 'tap profile tab' on OWN_PROFILE)

3. test_brain_live.py: Stochastic testing (5 runs, 60% min valid)
   to handle non-deterministic LLM behavior reliably

4. tests/integration/test_llm_provider_pipeline.py: NEW test layer
   mocking at HTTP level (requests.post) to exercise the FULL
   llm_provider → Brain pipeline. This would have caught the
   thinking substitution bug from day one.

Suite: 168 passed, 0 failed

GramAddict/core/llm_provider.py

@@ -287,6 +287,12 @@ def query_llm(
             req_data["images"] = images_b64
         if format_json:
             req_data["format"] = "json"
+        else:
+            # For free-text calls (Brain action extraction), explicitly disable
+            # thinking mode. Reasoning models like qwen3.5 put EVERYTHING in
+            # the thinking block and return response='', which is useless for
+            # action extraction. think=false forces a direct response.
+            req_data["think"] = False
 
         # Ollama passes configs inside 'options'
         if temperature is not None or max_tokens is not None:
@@ -349,13 +355,20 @@ def query_llm(
 
             logger.debug(f"DEBUG LLM PAYLOAD: response='{raw_response}', thinking='{raw_thinking}'")
 
-            content = raw_response or raw_thinking or ""
+            # CRITICAL: For free-text mode (format_json=False), do NOT substitute
+            # thinking for empty response. The thinking block is REASONING, not
+            # a decision. The Brain parser would extract random actions from it.
+            # For JSON mode (format_json=True), falling back to thinking IS correct
+            # because reasoning models may place structured output in the thinking block.
             if format_json:
+                content = raw_response or raw_thinking or ""
                 extracted = extract_json(content)
                 if not extracted:
                     logger.warning(f"Failed to extract JSON from content: {content[:100]}")
                 else:
                     content = extracted
+            else:
+                content = raw_response
 
             return {"response": content}
     except requests.exceptions.ConnectionError:

GramAddict/core/navigation/planner.py

@@ -94,6 +94,27 @@ class GoalPlanner:
                 logger.debug(f"🛡️ [Aversive Filter] Masking trapped action: '{action}'")
         available = safe_available
 
+        # 0b. No-Op Guard: Strip tab actions that navigate to the CURRENT screen.
+        # e.g. 'tap profile tab' on OWN_PROFILE is always a no-op.
+        noop_actions = set()
+        for action in available:
+            expected = ScreenTopology.expected_screen_for_action(action, screen_type)
+            if expected == screen_type:
+                noop_actions.add(action)
+                logger.debug(f"🛡️ [No-Op Guard] Stripping '{action}' — leads back to {screen_type.name}")
+
+        # Also strip actions where the HD Map says they go TO the current screen from OTHER screens
+        # (e.g., 'tap profile tab' isn't in OWN_PROFILE's transitions, but it goes to OWN_PROFILE from other screens)
+        for src_screen, transitions in ScreenTopology.TRANSITIONS.items():
+            if src_screen == screen_type:
+                continue  # We already handled this screen's own transitions
+            for action, dest in transitions.items():
+                if dest == screen_type and action in available:
+                    noop_actions.add(action)
+                    logger.debug(f"🛡️ [No-Op Guard] Stripping '{action}' — known to navigate to current {screen_type.name}")
+
+        available = [a for a in available if a not in noop_actions]
+
         # Build avoid_actions for HD Map route planning
         avoid_actions = (explored_nav_actions or set()).copy()
         if action_failures: