refactor: Fix code quality issues identified in analysis

1. Rename factory.py's EvoAgent data class to AgentConfig
   - Avoids naming conflict with base/evo_agent.py's EvoAgent

2. Export pipeline_runner functions in backend/core/__init__.py
   - Add create_agents, create_long_term_memory, stop_gateway

3. Consolidate PromptLoader to singleton pattern
   - Add get_prompt_loader() singleton function
   - Update all usages to use singleton

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

backend/core/pipeline.py

@@ -30,7 +30,7 @@ from backend.agents.team_pipeline_config import (
 from backend.agents import AnalystAgent
 from backend.agents.toolkit_factory import create_agent_toolkit
 from backend.agents.workspace_manager import WorkspaceManager
-from backend.agents.prompt_loader import PromptLoader
+from backend.agents.prompt_loader import get_prompt_loader
 from backend.llm.models import get_agent_formatter, get_agent_model
 from backend.config.constants import ANALYST_TYPES
 
@@ -1623,7 +1623,7 @@ class TradingPipeline:
 
         config_name = getattr(self.pm, "config", {}).get("config_name", "default")
         project_root = Path(__file__).resolve().parents[2]
-        personas = PromptLoader().load_yaml_config("analyst", "personas")
+        personas = get_prompt_loader().load_yaml_config("analyst", "personas")
         persona = personas.get(analyst_type, {})
         WorkspaceManager(project_root=project_root).ensure_agent_assets(
             config_name=config_name,