feat: add Chinese UI, Responses models, and mirrored Windows setup

Localize the web console and persist custom model and capability settings.
Use OpenAI Responses exclusively and move runtime limits into environment configuration.
Bootstrap uv and the virtual environment on Windows, support optional China mirrors,
and package source distributions without environments, build output, or credentials.

Update regression tests and repository quality checks for the new behavior.

artemis/services/token_meter.py

@@ -40,7 +40,7 @@ import threading
 from typing import Any
 
 from artemis.data_engine.context_vars import CURRENT_NODE_NAME
-from artemis.llm.google import usage_from_message
+from artemis.llm.usage import usage_from_message
 from artemis.utils.logger import get_logger
 
 logger = get_logger(__name__)
@@ -62,7 +62,7 @@ OPERATOR_NODE_NAMES = frozenset({"operator", "flashrunner"})
 def extract_usage(response: Any) -> dict[str, int] | None:
     """Extracts unified token usage from an LLM response message, best-effort.
 
-    Thin wrapper over :func:`artemis.llm.google.usage.usage_from_message`
+    Thin wrapper over :func:`artemis.llm.usage.usage_from_message`
     that keeps the meter's historical four-key shape. Returns None when the
     response carries no usable usage numbers.
     """