fix(api): forward real LLM usage in /v1/chat/completions response (#4310)

* fix(api): forward real LLM usage in /v1/chat/completions response

_chat_completion_response() hardcoded prompt_tokens/completion_tokens
to zero.  Now reads agent_loop._last_usage (set by process_direct
after every LLM call) and forwards the actual prompt/completion counts.

Streaming path is unchanged; usage is only surfaced in non-streaming
responses for now.

Fixes #4309

* fix: use defensive getattr for _last_usage and add it to all test mock agents

- Use getattr(agent_loop, '_last_usage', None) in server.py for safety
- Add _last_usage = {} to mock agents in test_api_attachment.py and test_api_stream.py
- Prevents AttributeError/500 when mock agents don't have the attribute

* fix(api): preserve provider total usage

---------

Co-authored-by: michaelxer <michaelxer@users.noreply.github.com>
Co-authored-by: Xubin Ren <52506698+Re-bin@users.noreply.github.com>

tests/test_openai_api.py

@@ -33,6 +33,7 @@ def _make_mock_agent(response_text: str = "mock response") -> MagicMock:
     agent.process_direct = AsyncMock(return_value=response_text)
     agent._connect_mcp = AsyncMock()
     agent.close_mcp = AsyncMock()
+    agent._last_usage = {"prompt_tokens": 100, "completion_tokens": 50}
     return agent
 
 
@@ -78,6 +79,25 @@ def test_chat_completion_response() -> None:
     assert result["choices"][0]["message"]["content"] == "hello world"
     assert result["choices"][0]["finish_reason"] == "stop"
     assert result["id"].startswith("chatcmpl-")
+    assert result["usage"]["prompt_tokens"] == 0
+    assert result["usage"]["completion_tokens"] == 0
+    assert result["usage"]["total_tokens"] == 0
+
+
+def test_chat_completion_response_with_usage() -> None:
+    usage = {"prompt_tokens": 150, "completion_tokens": 42}
+    result = _chat_completion_response("hello world", "test-model", usage)
+    assert result["usage"]["prompt_tokens"] == 150
+    assert result["usage"]["completion_tokens"] == 42
+    assert result["usage"]["total_tokens"] == 192
+
+
+def test_chat_completion_response_preserves_provider_total_usage() -> None:
+    usage = {"total_tokens": 77}
+    result = _chat_completion_response("hello world", "test-model", usage)
+    assert result["usage"]["prompt_tokens"] == 0
+    assert result["usage"]["completion_tokens"] == 0
+    assert result["usage"]["total_tokens"] == 77
 
 
 @pytest.mark.skipif(not HAS_AIOHTTP, reason="aiohttp not installed")
@@ -213,6 +233,7 @@ async def test_followup_requests_share_same_session_key(aiohttp_client) -> None:
     agent.process_direct = fake_process
     agent._connect_mcp = AsyncMock()
     agent.close_mcp = AsyncMock()
+    agent._last_usage = {}
 
     app = create_app(agent, model_name="m")
     client = await aiohttp_client(app)
@@ -250,6 +271,7 @@ async def test_fixed_session_requests_are_serialized(aiohttp_client) -> None:
     agent.process_direct = slow_process
     agent._connect_mcp = AsyncMock()
     agent.close_mcp = AsyncMock()
+    agent._last_usage = {}
 
     app = create_app(agent, model_name="m")
     client = await aiohttp_client(app)
@@ -364,6 +386,7 @@ async def test_empty_response_retry_then_success(aiohttp_client) -> None:
     agent.process_direct = sometimes_empty
     agent._connect_mcp = AsyncMock()
     agent.close_mcp = AsyncMock()
+    agent._last_usage = {}
 
     app = create_app(agent, model_name="m")
     client = await aiohttp_client(app)
@@ -393,6 +416,7 @@ async def test_empty_response_falls_back(aiohttp_client) -> None:
     agent.process_direct = always_empty
     agent._connect_mcp = AsyncMock()
     agent.close_mcp = AsyncMock()
+    agent._last_usage = {}
 
     app = create_app(agent, model_name="m")
     client = await aiohttp_client(app)