feat(llm): migrate from OpenRouter to OpusAgent

Replace OpenRouter (openai SDK) with OpusAgent REST API (httpx + polling).
LLMClient now creates topics per system prompt, submits async requests,
and polls for completion. Model tiers map to Claude model IDs
(opus-4-7, sonnet-4-6, haiku-4-5) configurable via env vars.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

scripts/run_phase1.py

@@ -159,8 +159,8 @@ def parse_args() -> argparse.Namespace:
         default=1,
         help=(
             "Numero di propose() LLM concorrenti per generazione (default 1 = "
-            "serial). 6-10 tipicamente accettati da OpenRouter qwen-2.5 senza "
-            "rate-limit; riduce wall time GA loop di 5-8x."
+            "serial). OpusAgent processa in coda FIFO; concurrency > 1 accoda "
+            "piu' richieste in parallelo."
         ),
     )
     return p.parse_args()
@@ -200,13 +200,13 @@ def main() -> None:
     print(f"OHLCV loaded: {len(ohlcv)} bars from {ohlcv.index[0]} to {ohlcv.index[-1]}")
 
     llm = LLMClient(
-        openrouter_api_key=settings.openrouter_api_key.get_secret_value(),
+        opus_agent_api_key=settings.opus_agent_api_key.get_secret_value(),
+        opus_agent_base_url=settings.opus_agent_base_url,
         model_tier_s=settings.llm_model_tier_s,
         model_tier_a=settings.llm_model_tier_a,
         model_tier_b=settings.llm_model_tier_b,
         model_tier_c=settings.llm_model_tier_c,
         model_tier_d=settings.llm_model_tier_d,
-        openrouter_base_url=settings.openrouter_base_url,
     )
 
     cfg = RunConfig(