Fix model config context/token limits and inject gutask orientation into agent identity

- Chat model max_tokens: 32000 → 4096 (was inflating context budget calculations)
- Utility model ctx_length: 8192 → 14000 (matches GLM flash actual loaded context)
- Utility model max_tokens: explicit 2048 cap (utility tasks need short responses)
- Utility model name: full zai-org/glm-4.7-flash to target correct LM Studio instance
- pull-agent-identity.py: include agent_id and gutask orientation block in system prompt
- gutask.md: expand lore command docs, note done auto-creates review record

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

agents/hermes/plugins/_model_config/config.json

@@ -11,7 +11,7 @@
     "rl_input": 0,
     "rl_output": 0,
     "kwargs": {
-      "max_tokens": 32000,
+      "max_tokens": 4096,
       "agent_id": 8
     },
     "max_embeds": 10
@@ -20,12 +20,14 @@
     "provider": "lm_studio",
     "name": "zai-org/glm-4.7-flash",
     "api_base": "http://host.docker.internal:1234/v1",
-    "ctx_length": 8192,
+    "ctx_length": 14000,
     "ctx_input": 0.7,
     "rl_requests": 0,
     "rl_input": 0,
     "rl_output": 0,
-    "kwargs": {}
+    "kwargs": {
+      "max_tokens": 2048
+    }
   },
   "embedding_model": {
     "provider": "huggingface",