Add Gemini-backed /api/generate-cards endpoint and update AI panel to use it

[CoderBrianNguyen]

Motivation

• Provide a server-side endpoint to generate flashcards from notes using Gemini models with robust parsing and retries.
• Improve reliability by falling back to local mock card generation when the Gemini API is unavailable or returns invalid responses.
• Surface clearer fallback/status messages to the client so users know when mock data is being shown.

Description

• Add app/api/generate-cards/route.ts which implements Gemini generation including prompt construction, multi-model attempts (gemini-2.5-flash, gemini-2.0-flash, gemini-1.5-flash), retry/backoff logic, response wrapper parsing, JSON extraction, and conversion into Flashcard objects, plus a GeminiRequestError type and fallback to mockGenerateCards.
• Implement robust JSON extraction with parseGeminiCards that can handle wrapped responses and malformed content and normalize difficulty values.
• Add retry delay calculation via getRetryDelayMs, request throttling handling (429/503), and capped exponential backoff with sleep.
• Update components/AIGenerationPanel.tsx to POST to POST /api/generate-cards instead of the old /api/gemini route and display server-provided fallback message when present.

Testing

• No automated tests were run as part of this change.

---

Codex Task