feat(backend): add optional Gemma 3n E4B AI provider

[danieleforte92]

Motivation

• Add optional support for the Gemma 3n E4B model (via an Ollama-compatible local runtime) so backend-only AI helpers can draft recipes, suggest ingredient categories, and explain previews.
• Keep PantryOps fully functional without AI by gating calls behind a feature flag and returning safe fallbacks when the provider is unavailable.
• Provide a minimal, non-mutative integration surface so UI work can be added later without changing core domain logic.

Description

• Added a small AI integration layer under backend/src/ai/ including aiProvider.ts (interface + errors), ollamaProvider.ts (generic Ollama-backed implementation), gemma3nProvider.ts (default provider wrapper), and aiService.ts (feature-flagged provider selection + fallback builder).
• Added light HTTP endpoints in backend/src/routes/ai.ts to validate input and proxy requests: POST /api/ai/recipe-draft, POST /api/ai/category-suggest, and POST /api/ai/explain-preview.
• Registered the AI routes in server.ts and included backend/src/ai/README.md documenting environment variables (AI_ENABLED, AI_PROVIDER, OLLAMA_BASE_URL, OLLAMA_TIMEOUT_MS, OLLAMA_MODEL), default behavior, and the endpoints.
• AI is disabled by default (AI_ENABLED=false), the service returns a 503 when disabled and a safe fallback response when provider calls fail; the integration never mutates stock/recipes by itself.

Testing

• No automated tests were executed as part of this change (bun test not run).

---

Codex Task

[danieleforte92]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Swish!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".