refactor: remove LiteLLM dependency — direct provider calls only

LiteLLM S7 proxy was a redundant middleman (94% endpoints unhealthy).
key_rotator.py already routes to 6 providers directly:
OpenRouter (7 keys), Groq (5), Kimi, OpenAI, Google, xAI

- key_rotator.py: remove litellm provider
- nba-agent.py: use call_llm from key_rotator
- nba-quant-daemon.py: use call_llm from key_rotator
- ingest-nba.py: remove dead LITELLM vars
- deploy.py: remove LITELLM secrets from S10 deploy
- S7 Space: PAUSED (no longer needed)

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

deploy.py

@@ -50,8 +50,7 @@ SECRETS = {
     "XAI_API_KEY": os.environ.get("XAI_API_KEY", ""),
     "COHERE_API_KEY": os.environ.get("COHERE_API_KEY", ""),
     "KIMI_API_KEY": os.environ.get("KIMI_API_KEY", ""),
-    "LITELLM_PROXY_URL": os.environ.get("LITELLM_PROXY_URL", ""),
-    "LITELLM_MASTER_KEY": os.environ.get("LITELLM_MASTER_KEY", ""),
+    # LiteLLM removed — direct provider calls only
     # ── Telegram ──
     "TELEGRAM_BOT_TOKEN": os.environ.get("TELEGRAM_BOT_TOKEN", ""),
     "ADMIN_TELEGRAM_ID": os.environ.get("ADMIN_TELEGRAM_ID", ""),