feat(costs): Stufe 2 — geschätzte Quellen (lokaler Strom, Cartesia, Notruf)

Lokales LLM: Call-Dauer gemessen -> Stromkosten je Job ((max-idle)*load*n/
3.6e6*Preis(EUR/kWh)*usd_per_eur), Kategorie llm_local. Cartesia-TTS: Zeichen x
cartesia_usd_per_char (tts). Notruf: pro Webhook-Alarm 0,20€*usd_per_eur
(alert), nur Doku, nie Sperre. Helfer in costs.py; neue Settings (GPU-Watt/Last/
Strompreis/usd_per_eur/cartesia/alert). Statistik: "exakt vs. geschätzt"
(Tooltip + Summenzeile, Kategorien tts/alert/llm_local). Tests: 318.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

app/providers/llm/local_openai_compatible.py

@@ -1,8 +1,10 @@
+import time
 from collections.abc import AsyncIterator
 
 import httpx
 
 from app.core import clock
+from app.core.costs import add_cost, electricity_usd
 from app.providers.llm.base import (
     LLMProvider,
     lang_instruction,
@@ -88,6 +90,7 @@ class LocalOpenAICompatibleLLM(LLMProvider):
         session_id: str | None = None,
         language: str | None = None,
     ) -> str:
+        t0 = time.monotonic()
         async with httpx.AsyncClient(timeout=120) as client:
             response = await client.post(
                 f"{self.base_url}/chat/completions",
@@ -96,7 +99,8 @@ class LocalOpenAICompatibleLLM(LLMProvider):
             )
             response.raise_for_status()
             data = response.json()
-            return data["choices"][0]["message"]["content"]
+        add_cost(electricity_usd(time.monotonic() - t0), "llm_local")
+        return data["choices"][0]["message"]["content"]
 
     async def stream(
         self,
@@ -106,6 +110,7 @@ class LocalOpenAICompatibleLLM(LLMProvider):
         language: str | None = None,
         **kwargs,  # z. B. on_tool_start — hier ignoriert (kein Tool-Calling)
     ) -> AsyncIterator[str]:
+        t0 = time.monotonic()
         async with httpx.AsyncClient(timeout=120) as client:
             async with client.stream(
                 "POST",
@@ -123,3 +128,4 @@ class LocalOpenAICompatibleLLM(LLMProvider):
                     delta = sse_delta(line)
                     if delta:
                         yield delta
+        add_cost(electricity_usd(time.monotonic() - t0), "llm_local")

app/providers/tts/cartesia.py

@@ -1,5 +1,6 @@
 import httpx
 
+from app.core.costs import add_cost, cartesia_usd
 from app.providers.tts.base import TTSProvider
 
 _API_URL = "https://api.cartesia.ai/tts/bytes"
@@ -139,4 +140,5 @@ class CartesiaTTSProvider(TTSProvider):
 
         if not response.content:
             raise RuntimeError("Cartesia TTS returned empty audio")
+        add_cost(cartesia_usd(len(text.strip())), "tts")   # geschätzt (Zeichen × Preis)
         return response.content