refactor(agent): make runner consume required runtime

tests/agent/test_runner_runtime_identity.py

@@ -4,25 +4,43 @@ import pytest
 
 from nanobot.agent.runner import AgentRunner, AgentRunSpec
 from nanobot.config.schema import AgentDefaults
-from nanobot.providers.base import LLMProvider, LLMResponse, ToolCallRequest
+from nanobot.providers.base import (
+    GenerationSettings,
+    LLMProvider,
+    LLMResponse,
+    ToolCallRequest,
+)
+from nanobot.utils.llm_runtime import LLMRuntime
 
 
 @pytest.mark.asyncio
-@pytest.mark.xfail(
-    strict=True,
-    reason="AgentRunner reads its mutable provider again between model iterations",
-)
 async def test_active_run_keeps_provider_captured_at_admission() -> None:
     first_provider = MagicMock(spec=LLMProvider)
     second_provider = MagicMock(spec=LLMProvider)
+    first_provider.generation = GenerationSettings(temperature=0.2, max_tokens=2048)
+    second_provider.generation = GenerationSettings(temperature=0.9, max_tokens=512)
     first_calls = 0
     second_calls = 0
-    runner = AgentRunner(first_provider)
+    request_temperatures: list[float] = []
+    selected_runtime = LLMRuntime.capture(
+        first_provider,
+        "captured-model",
+        context_window_tokens=16_384,
+    )
+    runner = AgentRunner()
 
-    async def first_chat(**_kwargs):
-        nonlocal first_calls
+    async def first_chat(**kwargs):
+        nonlocal first_calls, selected_runtime
         first_calls += 1
-        runner.provider = second_provider
+        request_temperatures.append(kwargs["temperature"])
+        selected_runtime = LLMRuntime.capture(
+            second_provider,
+            "future-model",
+            context_window_tokens=8192,
+        )
+        first_provider.generation = GenerationSettings(temperature=0.7, max_tokens=128)
+        if first_calls > 1:
+            return LLMResponse(content="done")
         return LLMResponse(
             content="working",
             tool_calls=[ToolCallRequest(id="call-1", name="read_file", arguments={})],
@@ -42,10 +60,12 @@ async def test_active_run_keeps_provider_captured_at_admission() -> None:
     await runner.run(AgentRunSpec(
         initial_messages=[{"role": "user", "content": "read it"}],
         tools=tools,
-        model="captured-model",
+        runtime=selected_runtime,
         max_iterations=2,
         max_tool_result_chars=AgentDefaults().max_tool_result_chars,
     ))
 
     assert first_calls == 2
     assert second_calls == 0
+    assert request_temperatures == [0.2, 0.2]
+    assert selected_runtime.provider is second_provider