refactor(run_agent): extract run_conversation to agent/conversation_loop.py

The 3,877-line run_conversation body — the agent loop itself — moves out
of run_agent.py into a dedicated module.  AIAgent.run_conversation is
now a thin forwarder that delegates to agent.conversation_loop.run_conversation
with the AIAgent instance as the first argument.

This is the largest single extraction in the run_agent.py refactor.
The body keeps all 163 self.X references intact (rewritten as agent.X),
all nested closures, all retry/backoff/compression machinery.  Symbols
that tests or callers patch on run_agent (_set_interrupt,
handle_function_call, AIAgent class attrs) are resolved through _ra()
inside the extracted module so the patch surface is preserved.

Five tests doing inspect.getsource(AIAgent.run_conversation) updated to
scan agent.conversation_loop.run_conversation. Two source-introspection
tests (TestMemoryNudgeCounterPersistence, TestMemoryProviderTurnStart)
updated to accept either self.X (legacy) or agent.X (extracted
form) in the matched assertions.

Live E2E verified on three model paths:
  * openai/gpt-5.4 (OpenAI chat completions via OpenRouter)
  * anthropic/claude-sonnet-4.6 (Anthropic Messages via OpenRouter)
  * moonshotai/kimi-k2-thinking (reasoning model, reasoning_content path)
Plus read_file tool execution, terminal tool, web_search.

tests/run_agent/ + tests/agent/: 4313 passed, 1 pre-existing failure
(test_auxiliary_client::test_custom_endpoint... — same as on main).

run_agent.py: 9800 -> 5944 lines (-3856).
Total reduction since baseline: 16083 -> 5944 (-10139, 63%).

tests/run_agent/test_jsondecodeerror_retryable.py

@@ -75,7 +75,9 @@ class TestAgentLoopSourceStillHasCarveOut:
     def test_run_agent_excludes_jsondecodeerror_from_local_validation(self):
         import run_agent
         import inspect
-        src = inspect.getsource(run_agent)
+        from agent import conversation_loop
+        # The body moved into agent/conversation_loop.py; scan both for safety.
+        src = inspect.getsource(run_agent) + inspect.getsource(conversation_loop)
         # The predicate we care about must reference json.JSONDecodeError
         # in its exclusion tuple. We check for the specific co-occurrence
         # rather than the literal string so harmless reformatting doesn't

tests/run_agent/test_memory_nudge_counter_hydration.py

@@ -120,10 +120,20 @@ def test_production_code_contains_hydration_block():
     """Smoke test: confirm the hydration code is actually wired into
     run_conversation(). If someone deletes it, tests above still pass
     against the inline replica — this fails them awake.
+
+    The body now lives in agent/conversation_loop.py after the
+    run_agent.py refactor; check both files for safety.
     """
     from pathlib import Path
-    src = Path(__file__).resolve().parents[2] / "run_agent.py"
-    content = src.read_text(encoding="utf-8")
+    repo = Path(__file__).resolve().parents[2]
+    src_ra = (repo / "run_agent.py").read_text(encoding="utf-8")
+    src_cl = (repo / "agent" / "conversation_loop.py").read_text(encoding="utf-8")
+    content = src_ra + src_cl
     # Anchor on the unique comment + the modulo line.
     assert "Hydrate per-session nudge counters from persisted history" in content
-    assert "self._turns_since_memory = prior_user_turns % self._memory_nudge_interval" in content
+    # The line uses ``self.`` in run_agent.py form and ``agent.`` in the
+    # extracted module, accept either.
+    assert (
+        "self._turns_since_memory = prior_user_turns % self._memory_nudge_interval" in content
+        or "agent._turns_since_memory = prior_user_turns % agent._memory_nudge_interval" in content
+    )

tests/run_agent/test_run_agent.py

@@ -5205,14 +5205,19 @@ class TestMemoryNudgeCounterPersistence:
     def test_counters_not_reset_in_preamble(self):
         """The run_conversation preamble must not zero the nudge counters."""
         import inspect
-        src = inspect.getsource(AIAgent.run_conversation)
+        from agent.conversation_loop import run_conversation as _rc
+        src = inspect.getsource(_rc)
         # The preamble resets many fields (retry counts, budget, etc.)
         # before the main loop. Find that reset block and verify our
         # counters aren't in it. The reset block ends at iteration_budget.
-        preamble_end = src.index("self.iteration_budget = IterationBudget")
+        # After the run_agent.py refactor the body uses ``agent.X`` instead
+        # of ``self.X``, so accept either form.
+        preamble_end = src.index("iteration_budget = IterationBudget")
         preamble = src[:preamble_end]
         assert "self._turns_since_memory = 0" not in preamble
         assert "self._iters_since_skill = 0" not in preamble
+        assert "agent._turns_since_memory = 0" not in preamble
+        assert "agent._iters_since_skill = 0" not in preamble
 
 
 class TestDeadRetryCode:
@@ -5220,7 +5225,8 @@ class TestDeadRetryCode:
 
     def test_no_unreachable_max_retries_after_backoff(self):
         import inspect
-        source = inspect.getsource(AIAgent.run_conversation)
+        from agent.conversation_loop import run_conversation as _rc
+        source = inspect.getsource(_rc)
         occurrences = source.count("if retry_count >= max_retries:")
         assert occurrences == 2, (
             f"Expected 2 occurrences of 'if retry_count >= max_retries:' "
@@ -5258,7 +5264,8 @@ class TestMemoryContextSanitization:
         a literal <memory-context> tag we don't silently delete their text.
         The streaming scrubber + plugin-side scrub cover real leak paths."""
         import inspect
-        src = inspect.getsource(AIAgent.run_conversation)
+        from agent.conversation_loop import run_conversation as _rc
+        src = inspect.getsource(_rc)
         assert "sanitize_context(user_message)" not in src
         assert "sanitize_context(persist_user_message)" not in src
 
@@ -5294,7 +5301,8 @@ class TestMemoryProviderTurnStart:
     def test_on_turn_start_called_before_prefetch(self):
         """Source-level check: on_turn_start appears before prefetch_all in run_conversation."""
         import inspect
-        src = inspect.getsource(AIAgent.run_conversation)
+        from agent.conversation_loop import run_conversation as _rc
+        src = inspect.getsource(_rc)
         # Find the actual method calls, not comments
         idx_turn_start = src.index(".on_turn_start(")
         idx_prefetch = src.index(".prefetch_all(")
@@ -5304,7 +5312,13 @@ class TestMemoryProviderTurnStart:
         )
 
     def test_on_turn_start_uses_user_turn_count(self):
-        """Source-level check: on_turn_start receives self._user_turn_count."""
+        """Source-level check: on_turn_start receives the user_turn_count."""
         import inspect
-        src = inspect.getsource(AIAgent.run_conversation)
-        assert "on_turn_start(self._user_turn_count" in src
+        from agent.conversation_loop import run_conversation as _rc
+        src = inspect.getsource(_rc)
+        # After the run_agent.py refactor the body uses ``agent.X`` instead
+        # of ``self.X``.  Accept either spelling.
+        assert (
+            "on_turn_start(self._user_turn_count" in src
+            or "on_turn_start(agent._user_turn_count" in src
+        )