fix: runtime deployment fixes for Docker Compose stack

- Add .gitignore for __pycache__, node_modules, .playwright-mcp
- Add CLAUDE.md project instructions
- docker-compose: remove host port exposure for internal services,
  remove Ollama container (use host), add CORS origin, bake
  NEXT_PUBLIC_API_URL at build time, run alembic migrations on
  gateway startup, add CPU-only torch pre-install
- gateway: add CORS middleware, graceful Slack degradation without
  bot token, fix None guard on slack_handler
- gateway pyproject: add aiohttp dependency for slack-bolt async
- llm-pool pyproject: install litellm from GitHub (removed from PyPI),
  enable hatch direct references
- portal: enable standalone output in next.config.ts
- Remove orphaned migration 003_phase2_audit_kb.py (renamed to 004)

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

packages/llm-pool/pyproject.toml

@@ -9,13 +9,15 @@ description = "LLM Backend Pool — LiteLLM router for Ollama, vLLM, OpenAI, Ant
 requires-python = ">=3.12"
 dependencies = [
     "konstruct-shared",
-    # Pinned: do NOT upgrade past 1.82.5 — a September 2025 OOM regression exists
-    # in later releases. Verify fix before bumping.
-    "litellm==1.82.5",
+    # litellm removed from PyPI — installing from GitHub
+    "litellm @ git+https://github.com/BerriAI/litellm.git",
     "fastapi[standard]>=0.115.0",
     "httpx>=0.28.0",
 ]
 
+[tool.hatch.metadata]
+allow-direct-references = true
+
 [tool.uv.sources]
 konstruct-shared = { workspace = true }