Add shared ai-service container as AI provider intermediary

All feature containers now POST messages to ai-service (port 8010) instead
of calling AI providers directly. ai-service routes to LM Studio, Ollama,
or Anthropic based on /config/ai_service_config.json. doc-service AI
providers removed; replaced by httpx ai_client.py. Backend settings
restructured to /api/settings/ai. Frontend gets dedicated AIAdminSettingsPage
and AI Service card in AppsPage.

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

features/doc-service/app/services/config_reader.py

@@ -1,18 +1,9 @@
 """
 Reads doc_service_config.json from the shared config volume.
-Caches the result for 30 seconds to avoid hitting the filesystem on every request.
-Uses asyncio.to_thread so the synchronous file read doesn't block the event loop.
+30-second TTL cache + env var overrides.
 
-Env var overrides (take precedence over the JSON config file, never committed):
-  AI_PROVIDER          — "lmstudio" | "ollama" | "anthropic"
-  LMSTUDIO_BASE_URL    — e.g. http://host.docker.internal:1234/v1
-  LMSTUDIO_API_KEY
-  LMSTUDIO_MODEL
-  OLLAMA_BASE_URL
-  OLLAMA_MODEL
-  OLLAMA_API_KEY
-  ANTHROPIC_API_KEY
-  ANTHROPIC_MODEL
+Env var overrides (all optional):
+  DOC_MAX_PDF_MB   — max upload size in megabytes (e.g. "50")
 """
 import asyncio
 import json
@@ -24,15 +15,6 @@ from pathlib import Path
 from app.core.config import settings
 
 _DEFAULT_CONFIG: dict = {
-    "ai": {
-        # Default: LM Studio running on the host machine at port 1234.
-        # Inside Docker, host.docker.internal resolves to the host; for local
-        # dev outside Docker use http://localhost:1234/v1 instead.
-        "provider": "lmstudio",
-        "anthropic": {"api_key": "", "model": "claude-haiku-4-5-20251001"},
-        "ollama": {"base_url": "http://host.docker.internal:11434/v1", "model": "llama3.2", "api_key": "ollama"},
-        "lmstudio": {"base_url": "http://host.docker.internal:1234/v1", "model": "local-model", "api_key": "lm-studio"},
-    },
     "documents": {"max_pdf_bytes": 20 * 1024 * 1024},
 }
 
@@ -52,43 +34,13 @@ def _read_config_sync() -> dict:
 
 
 def _apply_env_overrides(config: dict) -> dict:
-    """
-    Merge environment variable overrides into the config dict.
-    Env vars win over whatever is stored in the JSON file.
-    This lets the dev .env file pin the AI connection without writing to the
-    shared volume (which would affect all users).
-    """
     cfg = deepcopy(config)
-    ai = cfg.setdefault("ai", {})
-
-    if provider := os.environ.get("AI_PROVIDER"):
-        ai["provider"] = provider
-
-    # LM Studio
-    lms = ai.setdefault("lmstudio", {})
-    if v := os.environ.get("LMSTUDIO_BASE_URL"):
-        lms["base_url"] = v
-    if v := os.environ.get("LMSTUDIO_API_KEY"):
-        lms["api_key"] = v
-    if v := os.environ.get("LMSTUDIO_MODEL"):
-        lms["model"] = v
-
-    # Ollama
-    oll = ai.setdefault("ollama", {})
-    if v := os.environ.get("OLLAMA_BASE_URL"):
-        oll["base_url"] = v
-    if v := os.environ.get("OLLAMA_MODEL"):
-        oll["model"] = v
-    if v := os.environ.get("OLLAMA_API_KEY"):
-        oll["api_key"] = v
-
-    # Anthropic
-    ant = ai.setdefault("anthropic", {})
-    if v := os.environ.get("ANTHROPIC_API_KEY"):
-        ant["api_key"] = v
-    if v := os.environ.get("ANTHROPIC_MODEL"):
-        ant["model"] = v
-
+    docs = cfg.setdefault("documents", {})
+    if v := os.environ.get("DOC_MAX_PDF_MB"):
+        try:
+            docs["max_pdf_bytes"] = int(v) * 1024 * 1024
+        except ValueError:
+            pass
     return cfg