Deployment configs, CSS polish, and mock legal endpoints dressed as real integrations

8f3d355 adds a backend/Dockerfile (Node 20 Alpine, installs cairo/pango/libreoffice for PDF work, builds the TS backend, exposes 3001), a frontend/vercel.json, and ~41 lines of CSS: a glass-panel class using backdrop-filter: blur(16px), a bg-animated-gradient with a 15s keyframe cycle on a four-stop slate gradient, and a chatBubbleIn animation applied to each message on render.

The "Justice Canada MCP" path added to useCanadianChat in the same commit is a regex check - /act|code|legislation|statute/i - that, on match, returns a hardcoded Criminal Code, RSC 1985, c C-46, s 380(1) fraud provision. No MCP call, no network request.

8d446df adds render.yaml for a Render deployment, tweaks ChatView background and input area styling, and adds backend/src/routes/canlii.ts. That file is self-labeled MOCK CANLII INTEGRATION in a comment, and delivers: a /canlii/search endpoint returning two invented cases (2024 SCC 1, R. v. Mock Defendant and 2023 ONCA 555, Smith v. Jones) and a /canlii/document endpoint returning "This is a mocked primary source legal document for ${citation}". The TS LLM router gets fetchLegalCase and searchCanLII tools that call these mock routes internally.

The deployment configs (Dockerfile, vercel.json, render.yaml) are the only files in these commits worth evaluating for production use.