fix: expose embedding backend env vars in docker-compose and setup.sh

[ClaudioDrews]

Summary

The worker container (cognitive-worker) reads EMBEDDING_API_BASE, EMBEDDING_MODEL, and EMBEDDING_API_KEY from environment variables (see embedding.py), but docker-compose.yml never passed these through — only EMBEDDING_DIMS was exposed. Users wanting to use Ollama, vLLM, or any non-OpenRouter embedding backend in the Docker stack had no way to configure it.

Changes

File	Change
docker/docker-compose.yml	+3 env vars on worker.environment with backward-compatible defaults
setup.sh	+3 add_env calls for consistency with the existing EMBEDDING_DIMS entry

Default values

All defaults match the existing defaults in embedding.py, so current OpenRouter/Qwen3 users are unaffected:

• EMBEDDING_API_BASE → https://openrouter.ai/api/v1
• EMBEDDING_MODEL → qwen/qwen3-embedding-8b
• EMBEDDING_API_KEY → "" (optional, only needed for authenticated non-OpenRouter endpoints)

Verification

$ python3 -m py_compile docker/worker/services/embedding.py
✅ OK

$ cd docker && docker compose config | grep EMBEDDING
      EMBEDDING_API_BASE: https://openrouter.ai/api/v1
      EMBEDDING_API_KEY: ""
      EMBEDDING_DIMS: "4096"
      EMBEDDING_MODEL: qwen/qwen3-embedding-8b

Usage after this fix

# Ollama (local)
EMBEDDING_API_BASE=http://host.docker.internal:11434/v1 \
EMBEDDING_MODEL=nomic-embed-text \
docker compose up -d

# Or via .env:
echo "EMBEDDING_API_BASE=http://host.docker.internal:11434/v1" >> .env
echo "EMBEDDING_MODEL=nomic-embed-text" >> .env

Closes #29