Token tracking y cálculo de costes por mensaje

- Config: COST_PER_1M_INPUT y COST_PER_1M_OUTPUT configurables via .env
- OpenAI adapter: stream_options include_usage para capturar tokens reales
- base.py: acumula input/output tokens de cada iteración del agente
- planner.py: devuelve usage junto con el plan
- engine.py: suma tokens de planner + steps + review, calcula coste USD
- Response incluye usage{input_tokens, output_tokens} y total_cost_usd

Formato compatible con el bridge de Claude Code CLI para integración
con el frontend y reporting a Acai webservice.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/adapters/openai_adapter.py

@@ -44,6 +44,7 @@ class OpenAIAdapter(ModelAdapter):
             "temperature": config.temperature,
             "messages": messages,
             "stream": True,
+            "stream_options": {"include_usage": True},
         }
         if tools:
             kwargs["tools"] = self._format_tools(tools)
@@ -52,9 +53,22 @@ class OpenAIAdapter(ModelAdapter):
 
         tool_calls_acc: dict[int, dict[str, str]] = {}
 
+        final_usage: dict[str, int] = {}
+
         async for chunk in stream:
+            # With include_usage, the last chunk has usage but no choices
+            if chunk.usage:
+                final_usage = {
+                    "input_tokens": chunk.usage.prompt_tokens or 0,
+                    "output_tokens": chunk.usage.completion_tokens or 0,
+                }
+
             choice = chunk.choices[0] if chunk.choices else None
             if not choice:
+                # Usage-only chunk (last one with include_usage) — emit it
+                if final_usage:
+                    yield StreamChunk(usage=final_usage)
+                    final_usage = {}  # Only emit once
                 continue
 
             delta = choice.delta
@@ -99,16 +113,15 @@ class OpenAIAdapter(ModelAdapter):
                             tool_arguments=acc["arguments"],
                             finish_reason="tool_use",
                         )
+                    # Emit usage after tool_use chunks
+                    if final_usage:
+                        yield StreamChunk(usage=final_usage)
                 else:
                     yield StreamChunk(
                         finish_reason="end_turn"
                         if choice.finish_reason == "stop"
                         else choice.finish_reason,
-                        usage={
-                            "output_tokens": chunk.usage.completion_tokens
-                            if chunk.usage
-                            else 0
-                        },
+                        usage=final_usage,
                     )
 
     # ------------------------------------------------------------------