Make KB answer context size configurable
buildContext hardcoded a 2500-token cap on the source block sent to the LLM. On CPU inference the model reads every one of those tokens before emitting its first, which makes this the main latency knob after CHAT_MAX_TOKENS.
Expose it as CONTEXT_TOKENS (default 1800) and wire it through docker-compose.yml and .env.example.
Co-Authored-By: Claude Opus 5 noreply@anthropic.com