feat(agent): form pipeline + collection prompt workflow@20261021

- merge main: workflow/20261021 (collection prompt adds step-5 mandatory
  time readback-confirm before routing; non-motor/0003 branch rework;
  accident_cause unified schema), updated eval suites (photo_skip/
  button_guidance rewritten, no_false trimmed to asr, new
  non_motor_vehicle + time_confirmation with form/reply assertions),
  runner gains form/formMatches/replyMatches expectations
- prompts/collection.md regenerated from 20261021; extraction prompt
  archived (contentExtract description, 6-key whitelist)
- collect: extraction call per turn (extraction.md + current time/form
  injected), merge follows 更新表单 semantics (meaningful values only,
  number/string equality, diff -> turn_patch); accident_fields now uses
  external form keys
- backend: form_update flows to ChatResult and SSE formUpdate event
  (stream emits prefix, FormUpdate, body)
- fields: FORM_FIELDS whitelist with strict type validation
- verified: 155 unit tests; golden evals on real model - photo_skip 4/4,
  no_false 2/2, photo_button_guidance 12/12, non_motor_vehicle 14/14,
  collection_skip 10/12, time_confirmation 1/2; remaining 3 failures are
  model sequence/literal-template adherence, recorded in plan appendix

test/agent/test_collect_llm.py

@@ -85,7 +85,8 @@ async def test_collect_single_llm_call():
         lambda s, u: "<state>1002</state>请问有几辆汽车卷入了这次事故呢？"
     )
     result = await run_collect(collect_state(), llm)
-    assert len(llm.calls) == 1
+    # 对话调用 + 独立的字段提取调用（计划作业 4-4）
+    assert len(llm.calls) == 2
     assert result["response_text"].startswith("<state>1002</state>")
     assert result["turn_count"] == 2
 
@@ -98,6 +99,7 @@ async def test_collect_routes_to_photo_when_llm_says_so():
     )
     result = await run_collect(collect_state(), llm)
     assert result["stage_code"] == "2000"
+    assert len(llm.calls) == 2  # 对话 + 提取
 
 
 @pytest.mark.asyncio
@@ -113,7 +115,7 @@ async def test_collect_missing_prefix_repaired_once():
 
     llm = FakeLLMService(responder)
     result = await run_collect(collect_state(), llm)
-    assert len(llm.calls) == 2
+    assert len(llm.calls) == 3  # 对话 + 修复 + 提取
     assert result["response_text"].startswith("<state>1002</state>")
 
 
@@ -158,7 +160,7 @@ async def test_graph_collect_node_with_llm():
         config={"configurable": {"thread_id": "s1"}},
         context=GraphDeps(llm=llm),
     )
-    assert len(llm.calls) == 1
+    assert len(llm.calls) == 2  # collect 对话 + 提取
     assert result["response_text"].startswith("<state>1002</state>")
     assert result["stage_code"] == "1002"