feat: implement TeamRunner orchestration loop

Core orchestration engine:
- T1→T5 tier execution loop with dynamic task decomposition
- Config-driven adapter registry (team.yaml adapter_registry section)
- Config-driven runtime registry (top-level runtime section in team.yaml)
- Dynamic adapter loading via importlib (no hardcoded imports)
- Capability-based LLM dispatch (reasoning-heavy / capable / fast-cheap per tier)
- T4 defaults to coding_agent runtime (Claude Code)
- Blackboard integration for inter-tier state sharing
- Escalation handling with retry + tier promotion
- Dry-run mode for testing without LLM/VCS calls
- agent_personality injection from role_registry → system prompts

config/team.yaml

@@ -7,10 +7,21 @@ adapters:
   llm: anthropic
   vcs: github
   notify: openclaw
-  runtime: openclaw
+
+adapter_registry:
+  llm:
+    anthropic: "adapters.llm.anthropic:AnthropicAdapter"
+  vcs:
+    github: "adapters.vcs.github:GitHubAdapter"
+  notify:
+    openclaw: "adapters.notify.openclaw:OpenClawNotifyAdapter"
+  runtime:
+    openclaw: "adapters.runtime.openclaw:OpenClawRuntimeAdapter"
+    claude_code: "adapters.runtime.claude_code:ClaudeCodeRuntimeAdapter"
 
 models:
-  provider: anthropic
+  default_max_tokens: 4096
+  default_temperature: 0
   capability_map:
     reasoning-heavy:
       anthropic: claude-opus-4-6