feat(local-agent-v2): step 2+3 — unified runner + AgentConfig schema

Step 3 (prerequisite):
- app/schemas.py: add ContentTypeConfig + AgentConfig Pydantic models
- app/models.py: add agent_config (JSON, nullable) to LocalAgentConfig
- alembic migration a3b9c0d1e2f3: ADD COLUMN agent_config

Step 2 (runner refactor):
- Remove _classify_file() and _BATCH_FILE_CLASSIFIER_PROMPT (LLM classification step)
- Add Phase A: detect_content_type + preprocess (zero LLM, per file)
- Add _UNIFIED_PROCESSING_PROMPT (hot-swappable via Langfuse "unified_processing")
- Add helper functions: _format_projects, _format_metadata, _get_extraction_rules,
  _get_no_match_behavior
- Single LLM call per file with tools (classify + extract + create)
- Fix items_created: count create_* tool calls via _tool_calls_out param
- test_agent_runner_v2.py: 10 cases (2.1-2.10) with Langfuse eval scoring

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/schemas.py

@@ -273,6 +273,27 @@ class WsFloatingDomain(BaseModel):
     domain: WsDomain
 
 
+# ── Agent Config V2 ───────────────────────────────────────────────────
+
+
+class ContentTypeConfig(BaseModel):
+    """Per-type extraction config produced by the journey chatbot."""
+
+    id: str
+    label: str = ""
+    detection_hint: str = ""
+    preprocessing: str = "generic"  # handler name: "email_html", "plain_text", ...
+    extraction_prompt: str
+
+
+class AgentConfig(BaseModel):
+    """Structured agent configuration (replaces freeform prompt_template)."""
+
+    content_types: list[ContentTypeConfig] = []
+    global_rules: list[str] = []
+    data_types: list[str] = []
+
+
 # ── Agent Catalog ─────────────────────────────────────────────────────
 
 class AgentCatalogItem(BaseModel):